Thanks much Kris!
I understand that the exact scale/perf requires more intimate details about the problem domain.
Quick follow-up question:
" One session will in general be able to support quite a number of requests per second (see my advice on prototyping to get an idea in your specific case). If this is unsufficient, you can start looking at an architecture where multiple sessions are used to handle requests."
Do you think it would be useful if multiple sessions are loaded on the same machine or were you referring to loading these sessions on different physical boxes?