1) you can only provide useful names (not simply hashes) for things that are in directories, so a large number of bare files (easiest to update in IPFS) absolutely requires an external index linking the hash to a friendly name. I haven't found a suitable tool for automating any of this.
2) you can't modify an existing IPFS directory in-place and just get a new hash—you need the whole thing outside of IPFS to perform a CRUD operation. Maybe you can perform some trick with links or something, IDK, but it's definitely a pain. This makes One Giant Root Directory to Rule Them All very cumbersome.
3) IPNS is kind-of a solution, but centralizes things, and you can only host one of them per node-daemon (!) and last I checked these were still prone to availability problems and general quirky behavior.
4) Most software that doesn't speak IPFS natively (so, nearly everything) expects stuff to be in ordinary directories. Nothing expects to receive as config giant list of files that's changing all the time. They all expect one directory, or maybe a few directories, often with some reasonable structure. Translating between easiest-for-IPFS (a whole bunch of bare files, or smallish sets of closely related files each in their own root directory) and easiest-for-literally-all-other-software-that-uses-the-filesystem practically requires some kind of linking—I am not aware of any existing solution for this.
[EDIT] to be clear, (4) is no less a problem if you're using an IPFS FUSE mount. In fact you'd have to for linking to be viable at all, though I'm not 100% sure it works even then.