I doubt there's any real evidence or reasoning to show proof of work isn't feasible for email. Regardless, if you're sure the simple solution wouldn't work, you could try it with some increased complexity. The proof of work component could score posts based on a simple formula weighing proof of work against simple metrics like whether the text matches other recent posts and whether it includes any links. These scores wouldn't matter for npubs within a given web of trust. It doesn't make sense to give up on solving the problem just because you think the bare minimum solution isn't enough