That’s nice in theory, but if B2 has a castrophic disaster in thier one data center, that durability goes to zero.

That's just 8 9s. S3 has 11 9s.

Depends upon your client SLA then. You can also store the file twice on B2 for still less than the cost of S3 :)

Will failures be correlated then?

Then you should be comparing it to reduced redundancy S3, which is cheaper than regular S3.