Why it works
Against 3.Bc4 Black correctly grabbed on e4 and traded knights — so far the engine is on Black's side. 5...d6?? is where the position turns: it does nothing about 6.Ng5!, and after 6...Be6 7.Bxe6 fxe6 8.Qf3! both b7 and the e6/f-file complex hang at once. Here White's greed is simply correct: 9.Qxb7 h6 10.Qxa8 hxg5 11.Qxb8+! — the queen eats the rook, ignores the trapped-queen folklore and takes the second rook's defender too, emerging up a full rook. Every convention says the wandering queen must get punished; concretely, Black never gets one tempo to do it.
Refutation
5...f6! is the prophylaxis the position asks for: it takes g5 and e5 away from the knight before White's attack exists, and the engine even prefers Black (+0.7) — the 3.Bc4 line just isn't dangerous against accurate play. The Petroff rule survives contact: it is not taking on e4 that fails, it is forgetting that f7 needs one guard until the king castles.