Sure, the sizes of the communities is what matters (multiplied by the number of communities users on the server care about).
I think most of us are assuming larger instances are more likely to host the larger communities.
Actually, if I’m reading the protocol right, it’d be hard for a small server to host a highly active community anyway (for some value of highly active). So yes, some 2 person instance that was created to offload stuff could be the primary host for a massive community, but in practice it won’t.
If I’m reading the protocol right, it’s probably larger instances that will avoid more duplication, since:
I’m not sure I see where you see caching fitting in.
I am surprised I don’t see some kind of lower resolution digest concept in the protocol (which might be what you’re looking for)