There are multiple benefits from doing this.
Context switching is now obviously non-racy since there is only one process involved.
It should be somewhat faster since it doesn't involve a full host context switch, and it doesn't involve going through the host scheduler.
It will make a lot of UML code much cleaner and will allow throwing out a fair amount of code.
It will shrink the footprint of a UML on the host. There will be fewer kernel threads and task structures. Various process lists be shorter, speeding up searches.