You're right that ipfs is normally very resource hungry and slow. That's why we wrote our own implementation which used much less bandwidth and cpu and is faster (https://github.com/peergos/nabu). Be aware that we don't use Tor yet (see warning at beginning of book - https://book.peergos.org/).

Nabu is very minimal. It implements the structures and p2p protocols necessary to store and retrieve blocks and ipns mappings and do p2p mutually authenticated http requests, which is all we need for Peergos. Nabu itself doesn't have a concept of files, just blocks and their merkle links. The performance on a benchmark relevant to us is shown here: https://blog.ipfs.tech/2023-11-introducing-nabu/#benchmark