> there are only going to be a handful of test files Bingo. This used to be a well-known benchmarking cheat, because performance on a full (or even once-full)

> there are only going to be a handful of test files Bingo. This used to be a well-known benchmarking cheat, because performance on a full (or even once-full) filesystem can be very different than on a freshly made one. Almost as common as short-stroking, if you remember that. Anybody who wanted realistic results would run tests on a pre-aged filesystem. Cache warming is another example of a similar effect, and there are probably others across every area of computing. It's always really tempting to run tests in a maximally clean environment to get maximally consistent results, but that just doesn't reflect real-world usage.