Yes, I use 3-arity inequality functions with some regularity, much like PostgreSQL's BETWEEN operator, I've also seen it used by others. I think the most frequent use case for this is in tests for me personally but some of these are inside ares that can get large, so I would appreciate a performance optimization.
I would also agree that 4+ arity is exponentially rarer than 3-arity (I've never seen it in practice, but I assume there's been at least one person who has done (apply > coll) at some point to check if a collection is sorted), but I have no data on this beyond anecdotal experience.