Are there any QNX preferences to TCP server architecture?

I'm talking about a number of (remote) clients using TCP to make requests to a TCP server application that is listening 
on one TCP port, hosted on QNX.  There might be up to a couple of hundred TCP sockets active at a time (though it could 
be restricted to less).

The kind of preference I mean is, eg, is it better to have one thread handle each client connection or would it be 
better to use, say switch() to use one thread to listen for activity on all sessions?  Or some combination of the two.

Obviously there are costs/limits associated with each approach and there is a potential performance is with using 
switch().

Is  there an underlying message-based API that would be suitable at the application level?