I forgot where I read it, but at some point I read someone’s interesting observation that games are filled with stories of boys becomming men, and almost no stories of men realizing they’ve grown old. Especially as the average age of a gamer continues to work it’s way up there, I think the latter theme certainly will have an increasing place in the world of games.

Also, as someone mentioned Memento’s story being one of unraveling your own forgotten story, I at one point was working on a game with a story that worked like that. In fact, it was very disturbingly similar to Memento’s story in retrospect. Very similar ending twist. Only in mine, you were playing a computer that had deleted it’s own memory, rather than a person, but eh. Basically, think a hybrid of the film Memento and the game Paranoia (only you’re the computer rather than the people it controls) and that’s what it was sort of like.

Of course, at the time I couldn’t program to save my life, so it didn’t get anywhere. But hey, maybe I’ll make it someday.

As for Memento and the idea of a game being played backwards, of course there are games that flirt with the concept like Braid and Prince of Persia: Sands of Time, where the ending is actually the beginning. Or games like Majora's Mask or Planescape:Torment or Chrono Trigger that involve messing with the time stream to change the present (which I suppose could be closer to Groundhog Day than Memento, but they still have in common a non-linear story).

If you were to do this more explicitly in a game, you'd have to do it in a way that made the player really WANT to power down. Retiring is one way to do this. A downward spiral into madness would be another, darker path.

I don't see a problem with the player's skill being the inverse of the character's. This is essentially a metaphor for real life, isn't it? "It's a shame that youth is wasted on the young" and all that.

Personally, I believe the ’1/3 normal’ path would actually be detrimental to the experience. I would imagine it would not only further complicate the learning curve, by making it a learning spike. It would also require a clever narrative to prevent the player from feeling ‘cheated.’

They have to want to become weak. So seeing someone else become more powerful in your stead would work, but they have to be emotionally invested in that entity.

You could also have the first third of the game be “normal” and then the last two thirds being about losing your power. You could even structure a bunch of freeform missions where you play around with all your awesome powers, and then you get the option to begin campaign mode, which is far more linear and wherein you lose everything over time.

I think the game would have to be somewhat generic in its starting state: the game is interesting in its deconstruction of a pretty generic hero type, both psychologically and also in ability.

The problem is that your players become more adept at playing your game as they progress; they get more comfortable within the ergodic space you’ve created for them as they explore it. In doing this your player’s competence curve would be inverse of the accessible game space curve (how much freedom the player has to explore – keep in mind that more guns = more freedom in FPS games). Games have evolved player bootstrapping methods like tutorial levels to get players up-to-speed quickly. Eschewing this as a game designer would be equivalent to taking a BFG to your own foot.

Consider Portal. The level progression in Portal (some of the best ever devised) was the result of innumerable hours of play-testing. Imagine telling a player that they’re a Portal Device™ expert at the beginning of the game. They have just broken free of the laboratory and now need to break out of the complex in order to save themselves [and then the world?]. They start the game with GlaDOS telling them that there is no cake and that they should lie down and wait for the authorities to arrive. This wouldn’t work at all. You need to ride that learning curve.

In making a game like this, you’d be expecting your player to start the game as an expert with every tool [mechanic] in the game. Let’s take a look at the first-person genre. Each weapon is essentially a tool the player can use to help him/her explore the world you’ve created for them. HL2′s crowbar can be used to beat down enemies but it is more frequently used to break into crates and the like. The Gravity Gun and rocket launcher from the same game work similarly [isn't it amazing how those crates of rockets never run dry!?].

The Gravity Gun was new to the Half-Life universe in HL2. When it was introduced there was a short tutorial on its use. This ensured that the player continued on with at least some competence. The rest of the game was littered with Gravity Gun puzzles of increasing difficulty, culminating in its amazing use in the final level of Episode 2.

There is one way out of this: build this game as a part of a pre-established property. Valve could pull it off with Gordon Freeman in the game Half-Life 2, Episode 6: Denouement. At this point anyone playing the game should be familiar enough with the game’s mechanics and environments that they can start stripping Gordon Freeman of his tools. (Note that even Valve pulled the God of War strip-god-like-powers-from-the-player-early trick with the Super Gravity Gun in Episode 1.) However, this would have to happen with great care in order to ensure that the player would not become frustrated with the gameplay.

Outside of that you’d have to do a pretty generic genre game to get people into it. This because you’re presupposing expertise.

One last thing. You did mention the Batman-style, becoming less active and more tactical route. In creating a game like this you’d have to be extremely careful and deliberate with your gameplay changes; you could easily end up with a game that feels like four smaller, crappy games strung together with bubble gum. You could very easily end up with a reverse Spore and I don’t think that that would be very much fun for people.