How to explain poor performance on Xeon processors for a loop with both sequential copy and a scattered store?

Question

I stumbled upon a peculiar performance issue when running the following c++ code on some Intel Xeon processors: