The 'DoS' doesn't even require a protocol change to nullify. Indeed, there is a natural incentive already in the protocol that ensures it will never become a systemic problem. If large-time-to-verify-blocks ever became A Thing, miners will employ parallel validation. This will ensure that such large-time-to-verify-blocks will be orphaned by faster-to-verify-blocks.
Miners who gravitate to parallel validation will earn more income, and miners who do not employ parallel validation will become bankrupted over time. As will miners who create such DoS blocks.
This is already part of the protocol. No change is needed.
I've asked for a refreshment about 'parallel validation':
many miners currently mine empty blocks on top of unvalidated (but PoW-correct) new blocks. There's no reason to expect them to behave differently under BTU, so most miners would probably extend the chain with the high-validation-work block rather than create an alternative block at the same height.
Thus parallel validation doesn't get you anything unless a low-validation-work block is coincidentally produced at the same time as a high-validation-work block.
parallel validation only helps you in the rare case that there are two or more blockchains with the same PoW. Miners are disincentivized to create such chains since one of them is certain to lose, so the incentives probably favor them extending a high-validation-work block rather than creating a competing low-validation-work block.
Imagine block A is at the tip of the chain. Some miner than extends that chain with block B, which looks like it'll take a long time to verify. As a miner, you can either attempt to mine block C on top of block B, mining without validation but creating chain ABC that certainly has the most PoW. Or you can mine block B' that is part of chain AB' that will have less PoW than someone who creates chain ABC.
Harding's concern would be founded. But only to the point that
all miners would suddenly start performing
only zero-transaction block mining. Which of course is ludicrous.
What is not said, is that miners who perform zero-transaction mining do so only until they are able to validate the block that they are mining atop. Once they have validated that block, they modify the block that they are mining to include a load of transactions. They cannot include the load of transactions before validation, because until validated, they have no idea which transactions they need to exclude from the block they are mining. For if they mine a block that includes a transaction that was mined in a previous block, their block would be orphaned for invalidity.
So what would happen with parallel validation under such a scenario?
Miner A is mining at height N. As he is doing so, miner B solves a block that contains a aberrant quadratic-hash-time transaction (let us call this 'ADoS block' (attempted denial of service)) at height N, and propagates it to the network.
Miner A, who implements parallel validation and zero-transaction mining stops mining his height A block. He spawns a thread to start validating the ADoS block at height N. He starts mining a zero-transaction block at height N+1 atop ADoS.
Miner C solves a normal validation time block C at height N and propagates it to the network.
When Miner A receives block C, he spawns another thread to validate block C. He is still mining the zero-transaction block atop ADoS.
A short time thereafter, Miner A finishes validation of block C. ADoS is still not validated. So Miner A builds a new block at height N+1
atop block C, full of transactions, and switches to mining that.
From the perspective of Miner A, he has orphaned Miner B's ADoS block.
Miner A may or may not win round N+1. But statistically, he has a much greater chance to win round N+1 than any other miner that does not perform parallel validation. Indeed, until the ADoS block is fully validated, it is at risk of being orphaned.
The net result is that miners have a natural incentive to operate in this manner, as it assures them a statistical advantage in the case of ADoS blocks. So if Miner A does not win round N+1, another miner that implements parallel validation assuredly will. End result: ADoS is orphaned.
End result: Harding's concern is irrelevant. The quadratic hash time problem solves itself. No change to the protocol needed.