Implementation costs are pretty low - a small plugin or extra feature in each popular MUA, plus possibly a verifier in the MTA or MDA chain. Aunt Tillie's prehistoric Mac is taken care of by a plugin for Eudora v3, for example. Low-power mobile devices may have a problem generating hashcash themselves, but their upstream provider may be able to do it for them, especially if they already charge (however indirectly) for e-mail.So, OK, this is definitely going to turn off the non-hardcore spammers.
Those that are left will be the ones to send the law after. Theft of
service and computer intrusion on this kind of scale is a crime, no
matter whether the actual spam is.
What you also have to factor here is how much money will be saved in order to produce this reduction versus the costs required to convert everybody to hashcash.
If specialised hardware is built for one hash algorithm, it will be built for whatever other hash we migrate to as well, so unless the hash is actually cracked, we don't need to worry about changing it. The solution in the case of hardware acceleration being built is to move to a higher bit count, and start getting Taiwan to make cheap hash coprocessors so that ordinary users keep up with the spammers.You also need to factor in the ongoing costs to update the hashcash mechanism. You may need to either increase the number of bits collided, as generic hardware capability vs cost increases, or replace the scheme, if the hashing algorithm is broken, or specialized hardware is built.