Wow, I'm posting a lot more often than normal! I guess you can thank my new job at Linden Lab for that. :) We'll see how long this keeps up...
As far as I can tell, there are two types of games that people play for a long time: games that water down content and stretch it out through RPG grinding or FarmVille-style appointment systems, and games where you create. Okay, I guess there's a third - games with evergreen content complexity, like Triple Town, but these are very rare and I don't know of any very successful online games based on that principle.
World of Warcraft is the obvious example for games with lots of content, and even when it's watered down there is still a ton of it and it's very expensive to develop. FarmVille is an example that combines a thin layer of grinding with a thin layer of creativity. Minecraft is an example that combines a layer of grinding (harvesting resources) with a much deeper system of creation.
In all of these examples there is a social element as well, which is essential (even in Minecraft, minimal as it is), but even non-games have it (chat clients, forums, social networking sites) so I won't dwell on it here. Just keep in mind that the longest-lasting games tend to be social in some way.
In games with a creative element, like FarmVille and Minecraft, the grinding gameplay serves to give structure and ease the player into the creative play. As the player begins to tire of the grinding gameplay, the creative part is there to take up the slack. But the initial gameplay structure is essential to provide that hook and that ramp into the later experience.
In games where the creation experience is separate from the gameplay experience, there will be some players who only do creation and many more who only do gameplay. In this case you could set it up so the creators are providing gameplay content to the players. However, without watering the content down (with grinding), it is likely that the players will burn through content much faster than creators can create it. And because playing is separate from creation, most players who burn through gameplay will not transition to creation - they'll just leave.
I've imagined making a game where you create platformer levels like in N or Super Meat Boy, and earn points when other people play these levels and rate them highly. I still think that would be a cool idea, but I'm realizing that it would not work very well as an ongoing community experience. I doubt that anyone would spend months playing Super Meat Boy, as good as it is, while millions of people play games like FarmVille or World of Warcraft for a very long time. There's just not enough gameplay in a platformer to keep people going on level design alone.
But you could imagine a game where people create watery RPG content for other people, and where the creation and gameplay aspects are connected enough that there is a steady flow of players becoming creators. If you connect the creation and gameplay in a sloppy way, the two could collapse into each other, with people exploiting creation to farm gameplay progression (creating easy dungeons with gold everywhere and no monsters, for example). There could also be problems with finding appropriate content for players at various levels. But it should be possible to tune everything so it works as a self-sustaining ecosystem of playing and creating and moving between the two.
What would that look like?