Something I was also thinking about is whether the attack mitigation opens up new avenues for a malicious proposer to grief (publish late blocks which remain canonical).
I think the obvious case of a griefing proposer with 2 slots in a row is only marginally worse with the mitigation. They can prevent a re-org in slot i + 2 by publishing late in both slots i and i + 1. However previously they could also guarantee at least one late block remaining canonical by e.g. skipping slot i and proposing late in slot i + 1, which causes the single-slot condition checked by the proposer of i + 2 to fail (we only re-org when the parent block is 1 slot behind the block being reorged).