Naturally, the elite institutions should (hypothetically) be much more interested in PG applicants (who have extraordinary potential) rather than MG ones (who have "merely" great potential-- poor choice of words, perhaps, but you all know what I mean there, I hope).
This assumes ability maps more or less directly to potential. Potential is arguably a cocktail of personality traits, chance, connections, ability, and other factors.
Perhaps a better question is: potential for what? If the goal is an Earth-shattering new physics theorem, then E[PG]-E[MG] is probably > 0, with sigma[PG] > sigma[MG] to reflect the potentially deleterious impact of OEs, PG polymath indecision, etc. The university's admissions motive will reflect its risk tolerance wrt sigma.
However, let's not forget that the primary MO of universities is self-preservation, not novel scholarship. Novel scholarship is certainly an input into institutional longevity, but it is a bought resource and is probably only reputationally important below a critical level. If MG candidates are, on average, more financially successful in industry, then favoring MG over PG candidates is rational from an endowment management perspective.
