So there is no moving anything. Python compiled using pyenv for the Intel platform.
Which reports are you reading? Are they using C libraries or doing heavy math computation that is not CPU bound because there's no optimized ARM version yet?
What I am finding now is that python is working more than fine on M1, be it through rosetta or native code, which is consistent with your experience.
Also to clear things a bit, I parsed your "3.6 -> 3.9" as meaning that you had moved from Py36 on intel to Py39 on apple, which is why I asked if maybe there were speedups or differences from moving to 3.9.
I was unable to get all the versions I needed compiled natively, which means for now I will use the non-native versions.
I have tested Py3.9 compiled for arm64, and one particular test suite went from 9 seconds to 6 seconds. Other test suites were not as dramatic, especially larger ones, where it might go down from 40 seconds to 36 seconds.
Still a speed improvement, but not as dramatic as the 2017 MacBook Pro i7 -> 2020 MacBook Pro M1.