Isn't part of this that inserting into a btree index is more performant when the keys are increasing rather than being random? A random id will cause more re-balancing operations than always inserting at the end. Increasing ids are also more cache friendly