Why is Numpy slower on the vectorized for-loop of sigma clipping

Question

This question may have the same flavor as the one answered at Why is vectorized numpy code slower than for loops?

I want to implement a sigma-clipping method to rejec