I have wondered about that too. But when I think about a file system transaction, I immediately think about a file system transaction enlisting in an ambient transaction, together with a database transaction. This would make code much more simple/clean for cases where you must create a file and store metadata on the file inside the database. And on the other hand, some database systems do provide features for storing large file-like blob objects, which give you these transactional features.

So I think it depends on the context and what you wanna use it for. I don't see the transactional features of a file system as useful for actual users that are directly interact with files on their file system. It seems more useful in the context of applications that maintain files and where you likely do not want the user to directly interact with those files.