Most extant relational databases were originally designed pre-internet as a way to store and batch-process business records, rather than optimizing for things l

Most extant relational databases were originally designed pre-internet as a way to store and batch-process business records, rather than optimizing for things like single-record latency. For the batch processing use case, you usually want to be able to do range queries with the interior nodes of the tree in memory and the actual data ending up resident on disk. For the single-record lookup case (ie serving a user's data to them via a webpage) something like linear hashing actually does have better performance characteristics, although trees are usually good enough.