For distributed databases, when there are multiple conflicting transactions that touch the same data, I could see the proof of work as a method to determine which transactions win, especially in a global, multi-master model. When the proof of work puzzle is solved during conflicts, the transaction is accepted as truth and moves to a COMMITTED state. It seems kind of wasteful if you are paying for resources to solve challenges, but maybe there's a better way to decide who wins for trusted vs untrusted environments.

As an aside, Spanner I think acquires locks globally before writing. Maybe there's an interesting global-lock free model that just works using block chain, even if it's just for trusted parties (I.e. Google-internal).