I am unable to reproduce this bug. It's a timing thing (fails if something is too fast), so might fail if the machine is too fast, or if the accounting on CPU time is off due to other threads and cache misses not being counted against the right thread, or something like that. In any case, I've lowered the severity because this seems much more likely to be a test jig issue than an actual bug.
