The point is the account model. The implementation details of the relays is really a much more minor concern, given the protocol design. You can always route around bad actor relays! That creates much better incentives. Period.