There is always a small amount of variation, even on subsequent tests on the exact same machine. A 2 second difference is extremely small, statistically speaking. But this is indeed one of the reasons why you don't want a benchmark that is *too quick* to render. Variations of a few seconds are to be expected, even on the same machine, let alone on different machines of similar hardware, and of course on different hardware entirely. But imagine if the total benchmark render time was only 10 seconds. In that case a 2-3 second difference is 20-30%! Which, statistically speaking, is a huge difference. The problem in that scenario is that 2-3 second variations in render times are normal, so these variations have a bigger impact on shorter render times, making them less accurate for comparative purposes.