If there is concern about the block-with-valid-header-but-invalid-transactions-spam-attack, I have a strategy using sync flags that may drastically reduce the problem.

Sync flags documented here:

https://github.com/moral-agent/sync_flags/blob/master/README.md)

The strategy to defeat the above attack is illustrated here:

https://s32.postimg.org/e94tqdqat/sync_flag_invalid_block.png

The key is to relax the requirement that a flag commit to a completely valid block. The flag is valid if it commits to a valid block header, even if the block body is invalid.

From the perspective of an individual miner, they can safely commence mining a flag the moment they obtain (or discover) a valid block header.

As soon as the spam is discovered, miners can choose to either abandon the flag and return to mining on the previous block, or they can continue mining on the flag.

It's difficult for me to game out which of these strategies would be preferable. My first thought is that the miners should have the incentive to mine whichever option has the fewest miners, which should result in a 50/50 split.

However, the miners who continue mining the flag have a chance of ending up in a situation where they mine the flag before anyone mines a valid block. If this happens, it is sub-optimal for them. They can start mining for the next valid block but if someone else broadcasts a valid block header they will be in the same pickle that miners under the current protocol are: they must either keep mining for a valid block, or SPV mine the newly arrived block while they do validation. The third option, of mining a flag, is not available to them, because the flag has already been mined for this cycle.

As a result of the above, it may be most rational for miners to (upon learning that they are mining a flag on top of an invalid block) split their hashpower unevenly between the flag and continuing to mine for a valid block. The hashpower split reflects their estimates of the cost of the above negative outcome. I think the split would be pretty close to 50/50, but deviations from 50/50 would not necessarily be bad. For example, if they split 52/48, with more hashpower toward finding the valid block instead of the flag, then that decreases the likelyhood that the flag will be discovered before the next valid block, which is good for all of the miners. So it's a nice positive feedback.

*****

This approach mostly neutralizes the harm done by the (currently very rare) invalid block spam attack. As a kind of amazing side effect, the work done to produce the spam is incorporated into the blockchain cumulative Proof of Work, and the spammer is not paid for this contribution.