Nice writeup. A potential drawback of this is that random access to rows may slow down because of unbalanced b-tree indexes due to the most significant bits being largely the same. It may also have an effect on the distribution of load when sharding big tables across instances.