Gr. 7 with RC would just confuse things a whole lot.
Register changes need to happen on both scanlines and they're often time critical. Having big DMA one line and small DMA the next throws that timing requirement out so the same code couldn't be used on each scanline which would negate most of the memory saving by using lower resolution - in fact the overall memory usage would probably be more for less pleasing result and no CPU savings.
The CPU consumption really isn't an enormous hurdle. There's plenty of demos and a few games that use intensive screen DLIs/kernals and still leave sufficient time for the relevant processing to still take place.
Project M Wolf3D lookalike - uses Timer IRQs instead of DLIs which frees up some time but it's still a good example of a game engine which devotes significant CPU to creating a better screen layout but can still function as a game.
Possible hurdles I see mostly - PM graphics, we need some/most of these to be free, not letting RC use PMs would probably impact on the quality of image it can create.
- reserved playfield colours. I don't know if it's possible but having RC leave a particular PF colour alone so that it could be used for game objects would be desirable.