...but proper erasure encoding can make it many orders of magnitude more resistant.
Even on disk failure, some sectors can often be recovered with forensics. But then you need the storage providers' reputation at risk, so they have the economic incentive to pay for that forensics. Again it seems the Sybil attack is the problem, because they can blame the failure on a disposable Sybil.
Issue is not a full solution to a Sybil attack just Sybil resistance. When you get like 16 failures per trillion, its not really an issue. Even Amazon S3 has 15x more failures than that.
If in attacker has to store 10% of the entire network, but only has a 1.67e-11 chance of affecting a file I'd say thatis good enough. Worst case you can start adding economic incentives and disincentives. Look up the attackers funds/earnings on the blockchain for 3 months. "Oops lost your file out of 1 trillion. Here is $10k taken from the attacker."
My original criticism was that what have we accomplished that I couldn't just buy at Google's cloud.