I collected some more info on what I'm seeing on my 1.4GHz athlon:

I added some code to microtime() to save the last 4 sets of state
and to drop into ddb when the current result was before the previous
result.  the only part of the input state that was different
in the 4 sets was the LSB of the timer counter, and the four values
in chronological order were

	0x75
	0x6f
	0x68
	0x7c

I'm taking that to mean that the counter/timer hardware
that microtime() uses is unreliable in this box.
can anyone offer another interpretation?

for other people who see microtime() going backward,
there's probably a different cause, since this box has
the problem *much* worse than anyone else I've heard from so far.

-Chuck