I don't see what we lose if lead authors are forced to submit work themselves instead of going through the PI's account.

But yes, it'd be nice if arxiv also rate-limited co-author submissions with some higher number to discourage the most flagrant PI co-authorship abuses.