I do some sharding, mainly by grid-based bounding boxes. This allows for searching with an area or overlapping areas FTS index. It means there is some data duplication, but it is minimal compared to the speed boost.