Also, Freenet actively seeks out data from the network to store locally, to give plausible deniability to users (there's no way to tell if something is locally cached because the user requested it, or whether the client fetched it automatically). IIRC this can be disabled by limiting oneself to a whitelist of peers (preventing network chatter that might reveal Freenet's presence). IPFS will only cache what's requested, so it's more predictable and useful for building infrastructure (e.g. hosting assets for a normal Web site); and presumably leaner on data requirements too.
IPFS isn't really competing with Freenet IMHO; it's competing with HTTP and BitTorrent.
It has become a lot faster recently, try again :) When I look at its statistics during active usage it can easily show 1 MiB/s traffic.
You still can't expect HTML to load in sub-second delays, for large multi-MiB sites it may take a low single-digit amount of seconds, but there is an inherent cost to anonymity which probably induces a boundary on how fast things can be:
To achieve anonymity, data needs to be redirected across multiple people so the sender and recipient can't determine who each other of them is.
Redirecting stuff across a longer path than necessary slows things down.
>
> Redirecting stuff across a longer path than necessary slows things down.
That's skipping past a lot of details. There are still choices there. Freenet traded speed, efficiency and scalsbility for a specific security model. There are anonoyminity systems (e.g. tor) that are much more efficient but have different trade-offs.
I'm not sure why IPFS is more popular than Freenet, but the difference mentioned above certainly makes me more likely to run a IPFS node than Freenet.
Naive question: is there such a thing as Bittorrent over I2P or otherwise anonymized decentralized file sharing?
Yes. I2P is designed with P2P in mind. The default Java I2P client already comes with a bundled BitTorrent client, there is also a BitTorrent client called "Vuze" and is used by some non-anonymous users to help seeding a torrent across the clearnet and I2P simultaneously. Because of the nature of I2P, don't expect great speed. No seeder is "usual", 10 KiB/s is "normal" and 50 KiB/s with only 3 or 5 seeders is "good", overall, it feels as if it's still the early 2000s, but when you get used to it and adopt a "download and forget, come back in 3 days" mentality, it's not actually that bad. The fastest download I've ever seen was the movie Snowden, 20+ seeders and over 200 KiB/s+...
IPFS solved the data availability problem because you can incentivize storage miners to make your data available. This is something new, IMO.
Accessing it anonymously is possible with an extra protocol layer.