I was wondering for a while. Today, I reached the conclusion that I find it somehow a bit unsatisfying.
I get the rationale. Still, it is already half implemented (vassal X wants control of subvassal Z), it could be done fully. The vassal link should be tied to a list of territories. Yes, that mean you could have two suzerain. That could actually lead to interesting cause for wars. I do not think there is a necessity for an ultimate all-territories "top liege" whatsoever. So what: A is overlord to B, who is overlord to C, who is overlord to A. Well, it must be tied to titles: so A is overlord to B due to B being count of some part of A Kingdom; B is overlord to C due to C being duc of some part of B kingdom ; C is overlord to A due to A being baron of some part of C duchy.
What is the problem? In case of war, each can be involved or not, depending on the case.
We already have some strange business regarding laws that applies (kingdom law vs de jure). That would be must easier to apprehend: each territory has a visible proper hierarchy, unhomegenous, and law that applies at the one of the top liege of the territory.
Also, it would give much more sense to factions. Having a count part of a faction that is also king of something else would surely makes him much more annoying. That would clearly change the perspective of "important vassals". To the point that you could even think of dropping some titles just for the sake of getting rid of a vassal. Or, like in case of France vs England, make a fuss about the hommage ceremony to give reason to revoke.
The whole notion of revoking titles would be much more dynamic.
That could seem a bit more complex but that is not even sure, since it makes a few things more logical. That would tie kingdom politics to international politics, like they should.
(I admit AI might get lost - but from a human perspective, that would be better)