The main advantage IPFS has over torrents with magnet links is global deduplication / a global swarm. If you seed a torrent that just contains book A, and I seed a torrent that contains books A and B, I won't provide any bandwidth to people downloading book A through your torrent. If we were using IPFS, someone downloading book A will be able to pull from both you and me, even if we don't personally know about each other and started seeding independently.

Actually, BitTorrent v2 (as specified in BEP 52) has per-file hash trees, so it supports sharing a swarm if the same file is shared between different torrents. Not a whole lot of torrents use it though.

This sounds like a feature but it's actually a source of serious performance issues in IPFS.

More specifically, I think it’s basically the same idea as trackerless magnet links where you find peers for your file via a DHT. In fact, the Mainline DHT used by most BitTorrent clients is apparently very similar to the DHT used by IPFS. They are both Kademlia implementations.