> 5% would already be well within the margin of difference for separate identical clean installations of windows on the same hardware.

what is the source of this non-determinism?