T16875 suspiciously improves all the time on Windows
It appears that a number of recent MRs observe metric decreases in T16875:
It appears that none of the decreases really make it into master: https://grafana.gitlab.haskell.org/d/s3PVoM_Mz/ghc-testsuite-performance?orgId=2&var-environment=x86_64-windows-validate&var-test_name=compile_time%2Fbytes%20allocated%2FT16875%2Fnormal So it's just noise. Perhaps we should increase the acceptance ratio or otherwise figure out what's going on.