OK, there are tons of postings here for spam, so apologies if this has been done...
This is the simplest way to eliminate most spam that I can think of, so it probably has many holes in it.
1. A server with mail to send contacts the destination server and notifies that there is mail to be picked
up - very short message, provides mail message ID and server domain name only (mail1.abc.com or abc.com) and IP addresses are not allowed.
2. Destination server does either a local blacklist lookup or checks a public blacklist or both to see how many spams it has received from the sender server or domain, makes a decision whether to pick up or not based on its configured threshold. If no, then we are done.
3. If yes, destination server does a DNS lookup on the domain name and retrieves the correct message from the sender server using the message ID.
I would suggest the use of some encryption for the initial transaction to ensure that the message IDs are not intercepted in transit.
In the case of DNS abuse by registering tons of bogus domains, the registrars would have to be vigilant, but remember there would be a non-trivial cost in time and money to doing this.
This prevents zombies from sending mail direct from a PC, as the PC has no registered DNS address. It also provides an automatic blacklist system as large servers could publish lists of the largest spammer servers or domains.
So - why won't this work?