"Bad clients" surely affect the rest ... they degrade performance of whole sector because they consume larger than proportional air time leaving less air time (and hence capacity) for "good clients".
And that's true for any wireless technology (plain 802.11, nstreme, ...), it just might affect some more than others ...
wow but I thought that was fixed with NV2 protocol in the newest ROS versions.
NV2 protocol helps in such situations so that effect of a bad client on other clients is less.
Common misconception is that TDMA (time-division multiple access) is strictly round-robin (so that air time is allocated to each client regardless of needs). If that was so, the following scenario would be true:
let's say there are 10 clients in certain sector. 9 out of 10 are idle, only one wants to transfer data (e.g. running speedtest). TDMA allocates air time to all clients equally, so the active client can only consume 10% of available sector capacity. In reality, what TDMA actually means is that AP gives chance to every client to use some air time in a round-robin manner, but if client doesn't need it, AP will (quickly) poll the next client. In previous case TDMA would not waste 90% of air time, but perhaps 10%, so the (only active) client would be able to consume something like 90% of sector capacity.
And then it comes to the point of wireless transmission efficiency: TDMA allocates time slot to given client. How many bits of information is then transferred during that time depends on wireless link quality: if client has low signal strength and/or high level of interference, it'll use lowest possible speed (which includes robust modulation scheme such as BPSK instead of 64QAM, FEC with more redundant bits, ...) which means less data will get transferred during allocated air timeslot. The rest of data will wait in AP's Tx buffer. Now it depends on scheduling algorithm (implementation is not standardized and different vendors can implement different scheduling algorithms) ... but if it includes any kind of QoS awareness, then it might give higher priority (=allocate more air time) to clients whose data is waiting longer in the Tx buffer ... which means that bad client will receive higher priority than good clients and thus "steal" some of their air time.