I believe either could happen (gets slow, users start to drop).
This is not shared bandwidth the same way that, say, Ethernet is through a hub; the difference is how contention is handled.
In a hub, if two (or more) stations happen to transmit at the same time, they'll fall back, wait, try again. There's a formula, and some randomness is introduced so the next attempt by each "probably" will happen at different times (uses collision detection)
With wireless, the AP coordinates (by poll) all of the associated clients; the more clients associated, the longer the poll cycle. If there's some "dead air" time, a client can request time to transmit and has to wait for permission to proceed. (uses collision avoidance)
Thre are some other modifiers to the poll cycle and the transmission/response timing; generally speaking, no associated client can "speak out of turn" .... even to save the association. It'll time out, then re-associate.
If the system is still overloaded, the clients wait their turn, possibly, randomly time-out .... and occasionally drop.
While the APs usually can support a very large number of associations, in practice, the number of active clients any given AP can handle will be determined by how much of what kind of traffic is present, how much interference is in the environment, and some other factors (like processing power of the AP, memory, operating modes (802.11 b only, 802.11g only, mixed B/G modes, etc).
The Planet 3 book for studying CWAP (published by Osborne) has a pretty good description of the specific details ... check it out, it would be an excellent addition to any wireless library, even if you aren't studying for the Certified Wireless Analysis Professional certification.
Good Luck
Scott
