Still, on a business model where you only get paid a single time, cloud won't make sense as your other option is to do everything locally which is more profitable both in the long run and development-wise. Also, like you mention, there either has to be a fall-back, or there's the option that the game won't run at all if no such fallback is in place. Having to have a fall-back mode would increase development costs. And if you have a fallback mode, nothing on the cloud could be critical to gameplay (e.g. eye candy physics stuff that does not affect gameplay like some games that support HW accelerated physics), and if Cloud will be all eye-candy, there's less incentive to incorporate it that way due to continued costs (even though it will scale down like you mention in given time, its cost per player would be several times of a multiplayer game with dedicated servers,)
If Xbox One is Microsoft's final console, the cloud ambitions make more sense, after 10-15 years Xbox One could just act like a passive hub for streaming games, than it will be no different than Onlive where the whole game is processed on the cloud, which is a heck lot easier to achieve as you don't need to sync anything between local and cloud; as time passes by, your local resources will be small enough to be entirely neglected, because of the synchronization issues you would have to tackle with.