What's the reason for numpy that has different parallel rate under different Python environment?

Question

I have mainly used numpy and joblib to constructe a machine learning pkg. I developed it on Python3.5.x. The numpy mainly played a role in the com