[corty]

Maybe that is step 2. I agree that this would be hugely beneficial e.g. for databases where a low number of very fat replicas is preferable.