Just some quick testing you might be interested in. I might be able to add a few more hops to this in a week or so if you are interested...I've been curious about this myself
I'm posting this to show what you lose with hopping with single radio links, and not the actual bandwidth these particular radios can pass.
Boards are WRAPs@233MHz with CM9@36Mb and -70dB or better signals in the 5.3GHz band. The boards are downclocked from 266MHz to 233MHz (due to heat and the local climate) and the bandwidth does get reduced by a noticable amount at this slower speed.
[ test server ] -------- [AP1] <- - -> [AP2] <- - -> [AP3]
Bandwidth tests are with a Mikrotik on the test server (the bandwidth test client), which is wired to AP1 via 100Mb ethernet. The bandwidth server is the AP itself (which skews these results some). All 3 APs are bridged WDS with mode=ap-bridge with static entries. All OS versions are 2.8.28.
Speeds are based on values for the 10 second average after about 30 seconds. Ping times are based on a count of 25 pings. This network was idle when these tests were run.
Server <--- AP1 UDP: 27Mb, TCP: 21Mb (ethernet)
Server ---> AP1 UDP: 35Mb, TCP: 19Mb (ethernet)
Ping: min/avg/max = 1/1.2/2 ms
Server <--- AP2 UDP: 18Mb, TCP: 14Mb (1 Hop)
Server ---> AP2 UDP: 21Mb, TCP: 13Mb (1 Hop)
Ping: min/avg/max = 1/1.2/2 ms
Server <--- AP3 UDP: 10Mb, TCP: 9Mb (2 Hops)
Server ---> AP3 UDP: 10Mb, TCP: 9Mb (2 Hops)
Ping: min/avg/max = 2/2.0/3 ms
AP2 <--- AP3 UDP: 18Mb, TCP: 13Mb
AP2 ---> AP3 UDP: 17Mb, TCP: 14Mb
Ping: min/avg/max = 6/8.2/10 ms
I would expect you could nearly double these speeds if a Routerboard 500 or better were used. I ran some similar tests awhile back with Prism2.5 cards and noticed similar losses at each hop (after 5 hops, we had roughly 768K and 20ms pings IIRC).