well, to test response time you really need two systems -- the system
under test, and a second unloaded machine which is "pinging" some
particular subsystem to ensure it's still responding.

I'd be tempted to use "expect" driving something memory intensive like
emacs or a lisp interpreter over an ssh connection, timing how long it
takes to compute (+ 1 1) while some *other* job is running.

					- Bill