There's one more thing that the MR changes that needs addressing. We no longer use GetEnvironmentStringsW and depend on A->W conversion. It would be good to test what happens with variables that can't be correctly represented in ansi case. It's also interesting to check what happens when there are duplicated variable names due to conversion error.
I would also like to make sure it doesn't introduce regression in ucrtbase (assuming it handles it differently).