There is no reason to make a "hybrid" application for performance reasons, that is, a game that runs partially on local hardware and partially on remote infrastructure. You would have a distributed software system which is inherently more difficult to engineer and test. And you would combine expensive local hardware with expensive remote infrastructure. It doesn't make sense.
The only reasonable design for a cloud-based gaming infrastructure that scales its performance over time is to run games entirely in the cloud and stream their input/output to a generic thin client. This way, the game itself is not architecturally distributed and, hence, much easier to develop. Nevertheless, you would still benefit from the cloud's benefits like resource-efficiency and the outsourcing of upgrades and maintenance from customers to service providers.
There are, of course, features that you need a server infrastructure for, but these things are inherently in need of networking, for instance, synchronization of players in multiplayer games, social stuff, aggregation and analysis of data, etc. But nobody is "offloading" stuff that would otherwise run locally for performance reasons. Performance may benefit as a side-effect (e.g., a console does not have to run the server in multiplayer games) but the functionality itself would be inherently network-based.
This is true not just for technological reasons, but also for economic reasons: why would anyone pay for a cloud infrastructure to implement features that no player would care enough for to justify its price? Nobody would pay a monthly fee to run cloud-based servers that calculate pre-baked lighting. This is ridiculous. The developer would scale such features down such that they run on the local hardware. Its easier, and its free.
And finally, I haven't heard of any convincing idea for performance-motivated offloading of subsystems to a server. I have just seen some developers merely philosophizing about what might be technically possible without taking into account the necessary development costs or the actual impact of this idea. Everything else has been done long ago in online games.