This was a great change because these mechanisms can’t be effective if they’re in the node software [alone]. Custom node implementations could simply eschew that code. Instead, the solution must be at the network (protocol) level.
Perhaps if successfully storing and retrieving stored data increments the node’s reputation, which is a multiplier to payouts, it would provide an ungameable incentive to not overprovision? It would also be an incentive to longer running nodes, discouraging the overprovision → fill → delete-and-restart hack.