Why not? Maybe it is not possible currently, but what stops me from creating a graph databases optimised for these types of queries?

Currently a lot of research is being done on graph databases, and they can be potentially faster for queries that normally take a lot of joins.