I had forgotten about this, but by coincidence, it occurred again today, on 'x86 Windows7 3.7' after PR-14919 was merged. This may be the same machine as I might have left '7' off 'Windows' in the original report.
The 'windows timer' is used for time.sleep. From various web articles, I read that its resolution varies from 10 to 25, defaults to 15.8 milleseconds in Win7 (this is the figure I knew years ago), and defaults to 35 microseconds in Win 10. (Not exactly agreement.) time.sleep is hardly accurate for small fractions of a second delays, and delays may be much longer. I believe that root.after does better.
Zachery, what OS and version do you have?