With optimized RC (which deffer deallocation) you have no "constant-time performance guarantee", because you don't know at which time the deffered deallocation will happen... And all modern RC implementations are heavily optimized - especially in Apple's ecosystem...