My question was more for Zachary who proposed: "backporting GH-9901 to 2.7 would probably also be good even though it was solving a different issue".

But as I wrote, I'm not comfortable to backport such change in Python 2.7. I prefer to avoid changes except if we really have to fix it.

So yeah, let's see if the bug strikes back. At least, we should now see if the build systems attempt to kill python.exe or python_d.exe.