I think this is a really cool idea and it might make sense to use this for small collections (100k/200k documents).