My main point actually is that Apple don't pick this route for the fun of it. It HAS a need to be performant on mobiles devices (that have less luxury to get a GC that eat Ram) and that need represent A LOT OF MONEY. You can say any step about this, including making processors, are directly or indirectly related to the need.
P.D: I don't see Rc VS Gc as enemies, but as faces of the same coin. ARC is just a way to apply some Gc lessons to naive Rc. I think naive Rc is truly worse than Gc, but smart Rc is total fine and consider it the best of both in the general case...