> I ran pyperformance on my PR 4700. Differences of at least 5%: (...)

I tested again these 5 benchmarks were Python was slower with my PR. I ran these benchmarks manually on my laptop using CPU isolation. Result:

vstinner@apu$ python3 -m perf compare_to ref.json patch.json --table
Not significant (5): unpickle_pure_python; sqlite_synth; spectral_norm; pathlib; scimark_monte_carlo

Ok, that was expected: no significant difference.