This question is motivated by something from the set-theorist Hugh Woodin, a prediction he has made and styled as empirical, according to which a subtheory that he uses will not be shown inconsistent for centuries to come. That's a more "purely mathematical" example of this family of reasoning, though; what I have more specifically in mind right now seemed to come up in one of the essays I was reading about whether set theory has any special value in the production of physics theories.
But so it seems like you could make a trivial prediction like, "Someone will somehow find a way to use something from set theory, to come up with a new meta-explanation/grounding for physics." That wouldn't be a matter of falsity but vacuous success, then.
Perhaps a known(?) example of a better (less trivial) prediction, or something that could have been a prediction, along these lines, concerns the role of the Dehorney order in the background understanding of anyons. I mean, I don't actually know how much that order is involved in understanding anyons, maybe I'm misinterpreting the information I've looked over.
On the personal side of things, I have been trying for a few years now to work out a mathematical formalism for Lee Smolin's changing-laws-of-physics thesis. Offhand, I think I can see a way to model a physical universe in terms of well-founded sets of fields such that the "edges" of such a universe change (with spatial expansion and temporal progression) so as to "reprogram" the contents of spacetime for that universe. But I don't know if this will really "go anywhere," i.e. I don't know how to come up with a meaningful prediction on this basis. I would like to say, "I predict that we will be able to nontrivially define laws of physics in terms of infinitary logic and elementary embeddings, such that..." and go from there, but would that really be a scientific prediction, a meta-scientific prediction, or not really a "prediction" in much of a useful sense at all? Because at other times, when I think over the thousands of pages of set-theory material I've read, and contemplate the daunting content of nLab (over 18,000 entries! I hardly know where to begin with it), it seems as if there's probably no way that I myself will be able to find purchase in something that could not be easily adapted, like an all-purpose theory-engine oil, to make up whatever random "model" I'd be pleased to, and then it seems as if I'd be better off spending my time on something besides untestable conjectures about physics.