I just posted a simple mmtimer test for the regression tests.
Hmm. When I run it, I see only the expected behavior, both on a 2.4 and 2.6 kernel.
Note that the resolution setting is completely meaningless; the only setting that matters is the period.
And, as far as I can tell, it exactly reproduces the period (and I don't see any strange variance in the min and max).
By the way, shame on me for not having written such a nice regression test when I revamped this code. Thanks for covering for me.
Cheers,
Jer