Well usually ownership works by introducing an owner. For your example, you could keep the owned refs in a separate dedicated seq and have both your caches work on unowned refs. Then when the seq loses a ref you need to update the two Tables before and remove the corresponding unowned refs from it.
As usual, there are cons and benefits. The cost is that you need to have yet-another seq for the ownership, can be tedious. The benefit here is that you're nudged into a good design where your caches cannot keep the data alive unnecessarily (logical leak).
