I'll second these concerns. Time::HiRes may be up to snuff on Win32 (I don't see why it wouldn't be since it uses XS) but I'd read the source and put it through some torture tests to be sure. And you could always write your own XS code to handle timing if you needed to. But in addition, I'd be concerned about the accuracy of measuring keypress times on a standard keyboard
Incidentally, this looks like a good run-down of the commercial experimental psych packages, including E-Prime and others.