Sorry my bad, Tristan said in an email to me to use high proportional values, which, he meant you must use LOW proportional values :D. Now all is working.
The angular servo was fixed by using the inertia moment in the torque computation, see UPBGE commit 3d2e0c7faa3c92758bcc82e921e22d7bb37b3905.
Though you still have to set smaller values for smaller objects to get it stable.
Yes it will get a different result (though you were working on the middle of a release, not the most stable moment if there’s bug noticed as here), but the result should be better.