I don't see the point of hashing the key instead of picking the lexicographically smaller key. It will have the same effect and comparison is O(n), same as hashing.
Great question! You are correct, these methods would be equivalent w.r.t. tree "shape". The problem emerges downstream. In our shared database, we reply on the fact that user IDs are uniformly distributed. If we directly compared values, we start skewing the data towards the lower end of the range. By comparing hashes, we get the same heuristic effect without disrupting the uniform distribution.
