There was at least one company (Apollo, I think) that implemented demand paging on 68000 by using two 68000s. You had one, the leader, running as the "real" CPU, with the other, the follower, executing the same code on the same data but delayed by one instruction.

If the leader got a bus error they would generate an interrupt on the follower to stop it before it executed the bus erroring instruction.

The leader and follower would then switch roles, and the new leader could deal with the situation that had caused the bus error on the former leader.