I'd be interested to hear how you plan on reducing the mining variance with a nonoutsourceable proof of work.
Without a scheme to reduce the variance, nonoutsourceable puzzles would almost certainly lead to increased centralization (smaller miners effectively never solve the puzzle without collusion).
(I notice you're not calling for an immediate hard fork of bitcoin like Ittay Eyal and Emin Gün Sirer are)
Yes, it's absolutely essential that a switch to a nonoutsourceable puzzle also comes along with a variance reduction mechanism.
I have no clear winning solution for this part, but I've got a few ideas and have discussed them with Peter Todd, who also has ideas here. Basically the approach should be to internalize P2Pool, and to let miners choose a lower difficulty if they want (for a proportionally lower portion of the reward).
This is really subtle though. It would increase the amount of bandwidth needed. It would be easier to discourage / punish a block that has less difficulty. Also, some component of the reward should still be a high-variance jackpot... the nonoutsourceable puzzle is only effective when there's a luck component, and thus a big incentive for the real worker to defect.
So there's a ton more work to do. It's definitely not time for an immediate hard fork. But the discussion should begin.
Without a scheme to reduce the variance, nonoutsourceable puzzles would almost certainly lead to increased centralization (smaller miners effectively never solve the puzzle without collusion).
(I notice you're not calling for an immediate hard fork of bitcoin like Ittay Eyal and Emin Gün Sirer are)