Gabriel Kerneis and myself have just finished compiling some latency statistics for web servers. The results, assuming they are right, are rather surprising.
As we expected, processes are horrible and event-driven code is very good. The surprise is that kernel-threads, while unfair, are not as bad as we expected, and that user-space threads, if implemented right, are actually pretty good.
Did anyone do anything similar?

del.icio.us
Digg
Google bookmark
reddit
Simpy
StumbleUpon
Furl
Newsvine
Technorati
Tailrank