DontStop Personally I would say that MTE has "limitations", rather than "vulnerabilities". For sure it's not perfect, but other mechanisms that leverage a few bits per pointer, e.g., PAC, are imperfect too. CHERI would probably be substantially more powerful, but might cost more in hardware, and deployment effort would be nontrivial. In the absence of a free powerful technique, instead we have tradeoffs.
With respect to MTE, I think as long as code keeps getting MTE trips it means that people are still shipping vulnerabilities, that are each fairly likely exploitable. I think the GrapheneOS team has needed to exclude various vendor blobs from MTE because they trip, and there are plenty of forum posts about commercial apps tripping. To me that does not sound like we have entered a world where MTE is a waste of (fairly modest) resources.
Perhaps over the next week or so Google will clear up the situation.