I don't think that this is a good idea in pure mode. The whole point of pure mode is not to be infected by those kinds of environment variables. A quick grep of Nixpkgs shows dozens of packages that set LC_ALL, so there is plenty of evidence of LC_ALL having a material effect on build results.

From that point of view, HOME, USER and DISPLAY also sound like something that will affect the build.

In my experience the only effect LC_ALL seems to have is whether it's set or not (or sometimes whether it's set to a utf-8 locale or not)... Are there counterexamples?

There are quite a few python packages that depend on a UTF-8 locale for running tests, most of those are set because of that.

Is there a strong reason not to just set a UTF-8 locale in stdenv? Then it seems like respecting the user's LC_ALL would be much less of a divergence.

I wonder if it really is a python issue then... It should be able to default to the "C" locale - but for whatever reason that never happens. Anyway, I'm convinced this probably isn't the place to fix this issue.