I also wonder if it is problematic to have simple small query like that that can be design to make the database work endlessly?
No, because the "simple small" part of that doesn't matter. A user with access has various ways to affect denial-of-service on the server. If you don't trust someone from doing this kind of thing intentionally don't provide them a login. Accidental issues of this nature should be accounted for in other ways - like testing. And the process-oriented nature of the system helps too since CPU starving becomes difficulty (so memory ends up being the main concern).
The pathological case shown here is not all that concerning to me either.
That the implementation choice could have been different, and match other implementations, is interesting to me. But changing this kind of behavior is usually not an option. Though adding new features to accomplish the new behavior is something to consider.
