Random idea - as inspired by Honor+Intrigue and a thread on RPG.Net regarding the deadliness of systems, also probably some Dogs in the Vineyard as well, from what I recall.
Basically, everything comes down to the fact that, for me as a GM, killing a player's character isn't particularly interesting. If I want to kill a PC, it is going to happen. Unless they do their best not to engage with the world and plot, and then there's still always the "random mail bomb" if the game is modern, or "rocks fall, everybody dies". So what's the point?
Now, if I can get the players to the point of choosing to risk death, of making an active, positive affirmation that "this is worth dying for", then THAT I find most interesting.
What if the stake setting was based on how much the players were willing to risk and make it a decision that could be made. Basically, the design would consider of three tracks for the level of risk you were willing to accept - loss of your goal (Defeat), loss of your health (Injury), loss of your life (Death) - I quickly sketched out a diagram (Note: I'm NOT a graphic designer, or all that visual a person when it comes to graphics, so my apologies).
Possible problems I see with this - first, by leaving it in the players' hands, they may never get to kill/defeat the bad guy unless the players start losing. Of course, I could just have the same track for serious opponents who have the will to stick in the fight (which would make for a great demarcation for how tough the opposition is - lightweight minions are only willing to risk defeat, lieutenants are generally only willing to risk up to injury for their goals, and the true evil overlords are willing to risk their life to achieve something great.
A second problem is that this is a very narrative defeat - it requires the players and GM to come up with an answer as to how the defeat (of either side) emerges.
The third major problem that I see relates to the first - if the players are already losing, why should they bother to raise the stakes? Things are already going against them, so there needs to be some sort of mechanic that allows them some sort of bonus to success.
Benefits - I think the key is that it allows for results other than total defeat, loss, and destruction of a side in a conflict. Many games make it hard to retreat, so players learn to not bother, and GMs come up with more and more odd ways for leaders to escape death. Everyone may still need to come up with some explanation - but at least if everyone accepts the rules, they know this is going to be the end result. (See Problem #2)