The idea is very attractive in theory, but in practice proving time would be much more than 10 minutes inter-block delays.
What factors determine the proving time?
Can't it be parallelized somehow by proving different parts seperately, and then use these proofs as inputs in another SNARK-proof?