This argument is nullified by the fact that computing power continues to fall in cost, size and power usage. The user does not and should not care about where

This argument is nullified by the fact that computing power continues to fall in cost, size and power usage. The user does not and should not care about where his or her application logic is being run so long as this device is affordable. The vision you present adds questionable cost savings and substantial latency: (NY->SF->NY)/c will always take 32ms, best case. Taking this trip asynchronously instead will make sense for a very long time. Apple understands that shuffling data between devices is a real burden to users, so they've attacked that problem instead of chasing an esoteric one.