This is to some extent an implementation limitation, not a theoretical one.
Typical SQL databases support neither the data organization nor parallel orchestration features required to support these types of JOINs well. The practical issue is that you can't add these features to an existing database kernel architecture if it was not designed to make this feasible from day one, and people are rightly reluctant to design a new SQL database kernel architecture from scratch so that these features are available. SQL databases are trapped in a local minima.
Typical SQL databases support neither the data organization nor parallel orchestration features required to support these types of JOINs well. The practical issue is that you can't add these features to an existing database kernel architecture if it was not designed to make this feasible from day one, and people are rightly reluctant to design a new SQL database kernel architecture from scratch so that these features are available. SQL databases are trapped in a local minima.