Interesting. I'd assumed the performance degredation resulted from disabling the offending cpu optimization feature at the hardware/firmware level, which would seem to address the issue for anything on the box (vm and hypervisor).
But, perhaps that assumption was wrong and/or Amazon has another workaround.
