
Not sure how long you ran the benchmark for, or how quickly you ran it again afterwards, so I don't know how plausible this may be, but in addition to what was offered above, I would also check to see if anything funny is happening with heat throttling.