One major caution I have for you is that your talking about customer installs. Customers will not know enough about the interface to login and get signal levels nor will they know that a -50 is a better receive signal then -80. The problem is that you'll end up with lower signal levels then if you have a qualified technician installing the equipment. One or more bad connections will have a large negative effect on everyone on that AP. If you are planning to proceed with this then I highly recommend that you find a successful provider that is doing it and have a good long conversation with them to talk about their experience. I expect that you will find people receptive to talking about their success. If others are not having success doing this then take extreme caution. One good way to do this would be going to trade shows I've learned a lot from other WISP's at: (MUM's mum.mikrotik.com, WISPA events
http://www.wispa.org/, Animal Farm
http://www.afmug.com/)
Regarding good links, I would recommend this resource for you to ensure you know most of what is involved with establishing good connections:
http://mum.mikrotik.com/presentations/PL12/ahmad.pdf
Yes I believe you can pull off a 12x multiplier on an AP. You may need to employ QoS to keep low latency on important services like VoIP and to keep things like videos with constant streams but I defiantly believe this number can work. Keep in mind that 12x is a target and a maximum for an AP. Try to build to that but not exceed... a 10x is a good place to be 14x not so much however it used to be as high as 20x but that was before the days of Nexflix.
The larger the pool of customers using a shared resource the higher your multiple can be. If I run a 17x on an AP I will get complaints... If I run it on my up line connection that is the aggregate of thousands of customers I don't exceed 80% utilization. Consider a really small pool with a flat multiplier to see how it scales... consider a pool of 1 customer. Could you sell to one person 12Mbps and only purchase 1Mbps... or sell to two customers 12Mbps each and purchase 2Mbps... of course not so the multiplier scales with the size of the pool.
The network that I am getting my 12x from has two feeds of 250Mbps with an average peek load of about 60% to 80% per pipe and about 3500 customers. The most popular bandwidth package is 1.5Mbps although they range up to 20Mbps so I would assume the average is something like 2.5Mbps... This would be roughly 8750Mbps sold... that number divided by the bandwidth purchased is about 17.5x. At a size of 3500 customers I am sure we can exceed 20x.
This is a good resources that shows an example of managing bandwidth with QoS to offer good consistent connectivity for things that can consume resources like streaming videos. Implementing things like this would help you reach your budgets without having pore performance:
http://mum.mikrotik.com/presentations/US13/steve.pdf
Answer to Question 1: Simple network (I am reading "flat" layer 2 network): "Friends don't let Friends bridge networks".
I hate arguing because that's the way its done but one of the quickest and best answers I can think of is that none of the carriers use large flat Layer2 networks and when the offer Layer 2 connectivity its with MPLS. Your offering internet service and internet service is a web and is routed... Break the best practice model at your own risk (read 2am network emergencies and large support call volumes).
Layer 3 offers a lot of packet efficiency. For example you want to limit your broadcast traffic. If every time one client does a broadcast (like asking for an IP via DHCP) the broadcast would go from the station to the AP... then The AP would repeat it to everything on the bridge (All of the other Stations as well as the upline router). Then the upline router would repeat it to your other routers at the other tower sites and those routers would repeat it to the AP's and those AP's to the Stations... until finally everything in your network has heard that broadcast and ignored it except for your DHCP server which would answer through with unicast directly to the requester. Now this can be greatly reduced by using VLANS to separate each tower site into it's own Layer 2 network and you can also use client isolation on the AP and drop broadcast on the routers bridge. What your doing with this concept is basically creating tunnels between your AP and head end... at this point you should be thinking: why not just go all the way and create VPLS tunnels over MPLS between your end customer and core router (not a bad idea). Another big reason to have a routed network is that you cannot have Layer 2 loops (meaning redundancy) in your network without a headache of a Layer 2 routing protocol for loop avoidance or it will lock up your equipment. (You want routed loops with a nice link state routing protocol like OSPF because this avoids 2am network emergencies!)
The bottom line is that you don't want a large flat network. Route traffic to each site (OSPF), interconnect each site (create a web) with multiple paths for the traffic so that a power outage or failure of a single piece of equipment doesn't take half of your network down. Did I mention not having 2am emergencies... I was happy when those went away.
Answer to Question 2:
If you go with another vendor for equipment be sure to have some form of TDMA. The standard 802.11G is designed to operate where each client can "hear" the other clients. It waits to transmit until it hears an open time similar to a bunch of people in a room that can hear each other talking. If the clients begin talking over each other to the AP because they don't hear someone else talking then you will have collisions which is the AP distinguish between the clients... The solution is TDMA (Time Division Multiple Access) (read everyone having a scheduled time to talk). For successful outdoor wireless a scheduling mechanism is a must.
It is best to use both H & V on the same AP to reach the higher modulations. It would be possible to increase the number of AP's by creating a grid of H & V antennas... however offsetting an antenna by 90º only gives you 20db of separation. If your going to re-use a frequency on a tower you will need a lot of vertical separation as well or you will have to have the AP's in sync with one another (read Cambium with GPS sync). Without this the AP on the tower that is on the same frequency or on a near by frequency will be louder then any of the clients even if the antenna is pointed the opposite direction and turned 90º. ( TX: 20db - Polarization: 20db - FrontToBackRatio: 35db - very little Freesapce loss = could be as high as -40 depending on free space loss).
I recommend using MIMO 2x2 radios with NV2... set your channel width to 10MHz if your wanting to double the number of AP's on a tower... 3 non-overlapping channels becomes 6 non-overlaping channels... An array of 6 60º sector antennas is nice (get the best AP antennas you can). If you improve your RX & TX on one side by having a better antenna then your improving the entire link... With a better AP you improve every customer link with a better customer antenna you improve only that customer. Don't skimp on the AP. Also, increasing the passive gain of an antenna is much better then increasing the transmit power because increasing the antenna gain also increases the receive signal whereas the transmit power does not effect the Rx levels.
I think you still should give me a call. I highly recommend using a good consultant to help you build your network. It may cost a little extra money up front (there is a real chance it will save you money too) but I believe your return on investment will be very rapid. You don't just need someone with networking knowledge, you need someone that has worked with multiple networking models in the real world, that understands when something doesn't look right because they have fixed this kind of problem before. From the sounds of it you will learn all of this really quickly but there is a lot to learn and some of it is more art then science. Also you should consider systems integration early on. When working on support you need to get form installation notes to call history to usage graphs to billing etc fairly easily and quickly. The more information you have easily accessible the quicker and more accurately you'll be able to get a solution. Support costs are one of those hidden costs that can be real high or real low but that depends on how you build things. Furthermore you don't want to get to the point where you have to call in an expert and have them work on your network but they are lost on how everything is connected and where your issues are etc. Good design good planning and good record keeping will make your network need less maintenance and less time to repair issues when there is a need. One such example would be consistent and simple naming and addressing schemes. You don't want to talk to an consultant who is charging by the hour where you talk about the issue with Fred's tower and then mention Jan's tower and 10 minutes later you find out that the engineer thought those were two different towers and really they are just a married couple... Believe me bad habits die hard so be thoughtful from the beginning about scaling issues, keep good notes and records, keep things clean.