I'm largely in agreement with you, and I enjoyed your article on XTS. My general thought towards Elephant's contribution to CBC is that it forced coarse-grained manipulation, which is probably a lot more likely to crash a system as opposed to allowing meaningful attacks.

To risk a new design like Elephant implies that they found the attack model worth consideration; we know the rationale for including it, but we don't know the rationale for removing it. Perhaps knowing this would tell us something useful about current perceptions on disk encryption.

Lastly, I echo your disdain for sector-level encryption.

i may have to do some wikipedia editing later today ;-)