This was mentioned by joncnunn a few posts back. It's much better than my original #3 imo. There is a catch here though - you could very well use the gold you swindle from an AI to kill that very same AI. I'm sure if AIs could actually think they'd appreciate the irony.
Ofc the current repercussions of totally killing off an AI - namely the entire world turning against you with endless denounce and war spam - might well be enough to deter prospective Nigerian princes (so to speak). But sometimes with many civs around you can still find someone to trade with... And if the trick jump-starts your world conquest then the hostility doesn't really matter much. Some would term this a legit strategy, others would disagree and think it must be prevented altogether, not limited to one or a few uses. I'm not sure which camp I stand in at the moment.
I had thought about this, specifically with leaving one little nearly undefended city next to Alex or Monty, and letting them finish the job and take the diplo hit, as well as ending the deal.
But you could simply say all deals must go to completion, restarting at the end of a war, and if that civ is wiped out, that civ simply gets the lump of gold when they come back (or the remaining time on a luxury deal, if they are freed before the end of one). No one would benefit from the deal except that civ, if it is ever brought back though.
The logic is simple, at worst, a civ has one city with 1 pop that has survived. We know that you can liberate cities for "dead" civs, as well. They still exist, but presumably as a puppet or annexed state of at least one other civ. Even if they are eliminated, one could still give those peoples the rest of the deal, even if that civ has lost all it's cities.
As the people still exist, it's logical that other civs would expect you to keep paying you debt to those people, no matter if they are a major civ or a puppet state.
You could still potentially stay at war with a defeated enemy AND try to protect his last poor cities from enemies, but there are very few situations in which the long term investment put into protecting them at the end of the game would be worth the upfront gain, unless you've really pretty well already won.