First: I don't think Matthew Green "resorted" to anything. I think he chose the blackmail example because it's easy to understand on a personal level: we all use repudiatable protocols in other contexts (like Signal), so why wouldn't we want it on our emails?

Second: That's not how blackmail works. It's contingent on what the extorted party thinks, not the cryptographic integrity of the blackmail material. That's why mass blackmail spam campaigns (that DKIM fails to prevent, ironically enough) are remarkably effective. Publishing DKIM secret keys after their expiry doesn't magically prevent blackmail; it just removes one more tool from the blackmailer's toolbelt for instilling fear in the target.

The problem is that it's not clear about it. It looks like it does quite a bit, and the counter-argument boils down to "someone could have cracked the secret key", which everyone always is told is the thing that is impossible. So you get plenty people believing and claiming DKIM can do that. This would be fixed by obviously breaking it.