Without benchmarking it, my first instinct would still be to design the schema in a way to avoid recursive queries, if possible. I'm thinking of stuff like a si

Without benchmarking it, my first instinct would still be to design the schema in a way to avoid recursive queries, if possible. I'm thinking of stuff like a simple tree, e.g. categories with subcategories arbitrarily nested. In that case I'd probably just store all the parents as well, not just the immediate category. That is more work in the insert/update case, but makes queries that include all children trivial.