I mostly adhere to the rule of thumb "if it ain't broke, don't fix
it". So I would only endorse such changes if they address actual pain,
rather than possible pain.

Note that that rule of thumb was born out of worry about another
possible pain: the observation that some fraction of well-intentioned
fixes actually break something unanticipated. An (imagined) example in
this case: if someone mocks time.time() in a unit test, your fix
*might* break their test.