https://bugs.winehq.org/show_bug.cgi?id=55786
--- Comment #10 from Esme Povirk madewokherd@gmail.com --- Turns out I forgot to update the index of the median, so I was actually getting something very close to the minimum when I increased the sample count. This makes much more sense of my results and gives me results that are (mostly) closer to what I'd expect. In particular, I'm getting timings both greater and less than 10 ms on Wine, instead of consistently less.
I've *also* started getting wonky values again from the Windows VM's. The fact that this didn't happen when using minimums suggests to me that 15 ms is generally accurate on Windows machines, and the values closer to 20 ms were errors.
I think I'll see if increasing sample size helps on the Windows VM's, and if not I'll just stick with a large range like we already had.