So the issue is that the fix for
JENKINS-47928 was too heavy a hammer - it didn't call post for a parallel parent if there wasn't an error specific to that parallel parent (or its children, obviously). That means it wasn't run for cases where no failure had been encountered at all...which was not ideal. =)
So I think I've fixed it by tracking when a parallel parent is skipped due to earlier error/unstable/when and not running the post if the parent was skipped. This doesn't do anything about some other similarish cases - i.e., if one of your parallel child stages fails before another parallel child stage completes successfully, the post failure will stick in for the successful child stage, since the build as a whole is a failure by then - but that will have to wait for more granular status controls.