You'd propose a risky change that moots millions of dollars of deployed hardware to address a theoretical implementation of a theoretical weakness?

...A implementation which requires hashers to voluntarily sign up for using services which hide extra data from them? That that an attacker wants to hide because it doesn't want hashers to know that they're beginning to perform the aforementioned theoretical attack (which can't really be hidden once its actually happening actively, as miners will see their solutions being delayed and the network will see orphans). ... and yet hashers are somehow not going to be smart enough to realize that anyone asking them to switch software/firmware is doing so because they want to perform that attack?

Existing mining software (BFGMiner) already validates the sanity of prevblocks in requested work, e.g. validates that a pool isn't asking the miner to fork its own past work. I don't think there is any risk of anyone sneaking this in on anyone.