I’m not sure I understand the bit about time-ordered values being a better fit for B-trees. Actually, I’m sure that I don’t understand it... If anything, I’d naively assume that inserting already-sorted values would be a worst case, always going down the same branch of the tree and triggering re-balancing operations left and right (figuratively — literally mostly left!). The Wikipedia article on B-trees notices that there are some shortcuts that can be taken if one knows one is going to do a batch insert of sorted values. Is that what happens? But then how does Postgres know this?
Btrees don't need to be rebalanced. They only split. The cost of splitting to the root is counteracted by the cache locality of always accessing the same set of pages. Inserting into the last page only needs O(logn) page accesses and O(logn) page writes worst case.
