The current system's problem is probably caused by mysql's row
locking. Simply switching to a versioning-based database should make
things much better.
As about maximum load possible, I have one AOLserver running a
custom Tcl script as the request processor. Each hit requires 2 db
queries against a table with 500000 rows in it, plus incrementing
about 5 nsv variables (which implies alot of locks taken). The thing
does an average of 31 hits/second, with peaks jumping over 50/second.
The db hit average is around 15/second, much lower than the expected
62/second, since the script uses caching (nscache).