I can attest to the usefulness of Filecoin. But I do think you are right in that IPFS is for immutable file/asset storage, and not as efficient for mutable data. David, one of the core teams at IPFS, reached out to me the other week about how CRDTs on top of IPFS can fix that. That is what we have solved, mutable graph data that can update in realtime or offline, across a ad-hoc mesh network.
IPFS is still critical here though, because using our Radix Storage Engine (just released) adapter, we can batch high volume mutable data to disk (aka IPFS) for storage and retrieval, which once fetched gun kicks back into gear and does sync.
Ethereum is just not at this scale yet, last I heard it was doing about 7 transactions per second (and Bitcoin doing 4). We're doing 2K end-to-end (across a 4 peer network hop) table inserts/second! I demo this on stage here: https://youtu.be/nTbUCTgLmkY
In that talk, I also explain how to do load testing across a distributed system in order to verify its correctness, aka do simulations on failures like with Kyle Kingsbury's (Aphyr) Jepsen Tests.
So while Filecoin isn't ready yet, its market cap represents a massive emerging opportunity. If Uber unicorn-ed crowdsourced transportation, imagine what a truly P2P incentivized crowdsourced storage system could do. Remember, all machines need electricity (Bitcoin), computation (Ethereum), storage (filecoin), and bandwidth (what we're tackling, with data sync, and incentives).
If you look at it in that context, it makes sense. Hopefully the crypto-P2P community will deliver :) to keep up with the demand. Cheers!