I agree, that if you don’t need to rebuild the deformation part, it can be a big gain to keep that part in a Maya asset while changing the behavior control part.
But, at less for the project that I work, where everything is rigged and animated manually (Cloth, hair, chains, and other character details) the bast majority of changes comes from adjusting the controls and deformation of these elements. So there is no easy way to keep the deformation part.
Probably it is possible in some situations, but other issues can easily arise from this, like leftover nodes or untracked modifications
Going the route of full rebuild each for each rig iteration ensures that everything is more under control, and when we scale to many rigs (we have projects 60+ characters and several resolutions, that can be hundreds of rig assets) having everything serialized and sharing many of the data allow us to do mass updates just changing a few data pieces.
But just to be clear the idea of having the behavior and deformation rig separated (this is very common also in games) is a good idea and not necessarily exclude one to another. But for my experience, it doesn’t solve all the cases that you can find in production.
In the mGear roadmap, I have planned a tool to improve the way that we use custom steps. It will be a nodal processing graph. In this tool, one of the functionalities that I would like to have is the ability to consolidate part of the execution graph in a Maya asset, so you can continue the build from that point or spawn different branches for different character variations. I hope with this approach we can get the best of the 2 approaches