I've not been so concerned about index sizes. Blobs are an order of magnitude bigger problem space with I think. I think figuring out how to make indexing faster OR to find good times to do particular indexing, and when to allow queries through without waiting for the indexing to complere would be important.
OH, that was an idea I wanted to inject @dominic - the ability to make a query and say "just give it to me what you got now, I don't want to wait for the indexes".
example use cases : I want to see posts, but I don't really really need the updated names and faces of everyone to read that content.