If I'm reading the protocol right, it's probably larger instances that will avoid more duplication, since:
- There's a higher chance they're going to have more communities shared among users (for really tiny instances you're probably going to get a lot of overlap since those people likely have interconnected interests, but I expect that would fall off quickly, but then converge at scale).
- The larger number of users will mean they 'use' more of the content they're pulling down (I can't read all of a highly active community in a day, but 1000 people together checking through the day might 'use' it all).
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)
At one point he posted a response that started with “A:” before editing it to drop the A:
which made it clear he was posting canned answers at the very least