Player's brains are rather remarkable at turning crap into content. It's like magic. A good GM is also good at this, and the players and the GM can basically toss crap into a pot, stir it with their brains, and out comes a story.
If you're running a game, as a GM, you really only have two responsibilities: controlling pacing and providing new data.
Really, my specialty has always been in providing new data. Not only do I provide interesting data, I also have the ability to provide the right data to cause a cascade of players assembling data into something fun and useful. There's no better sight than to see a player's eyes bug out as he solves a piece of the mystery.
But the thing is, I think it's not actually a very important skill. First, since it really can't be automated for use in computer games, it just doesn't have the adaptability I wish it did. Second, players are so good at making structure out of crap that they really don't need such carefully considered data. Sure, higher quality ingredients make for better food, but playing is as much about enjoying the cooking as the eating.
Yes, it's possible to structure a game such that all the content you provide is structured such that players put it together in a specific way, at which point the next content you provide is simply what makes sense. This is called "telling a story". The pacing and progression of the story is so clearly defined that players simply can't accidentally wander off and require you to change your plans to fit their ideas. Most games are written like this, or in a similar way, where players will be in the right mindset when they choose to proceed (see Grand Theft Auto III).
There are about a million downsides to this, but perhaps the strongest issue is that it is almost impossible to have a multiplayer game set up with this kind of predictability. Players will start generating their own content immediately, talking with each other and doing some role playing. This will destroy not only the pacing you've got in mind, but also the story itself. The story won't connect with them strongly enough, because they've deviated from their set roles in it. (I've known several GMs who had this problem chronically - I think most GMs have it from time to time, when they grow to "love" a particular idea.)
However, when you have a group of players, they are happy to make their own content and incorporate it into the content you feed them. And, as I said, you can literally feed them random crap and use a simple feedback system to feed them more of the crap they seem to like. Eventually, a plot will develop even if none was intended. This apparently works especially well in pilot episodes of Prime Time, at least the ones I've participated in. Sure, in the end, it doesn't make a whole lot of sense and isn't very cohesive. But, damn, it was fun.
(I am going somewhere with this, as unlikely as it seems...)
Generally speaking, groups of three to five have the best success in combining content to make sense. If you have more players than that, it is generally best to split them into sections, even if they are running at the same time. Most successful LARPs do this - creating little "interest clusters" of three to five players, with one or two players who are interested in tying interest clusters together.
But the reason it works is the extremely high bandwidth of face-to-face conversation. There's a lot of data to be processed: the more pieces, the smaller they are, and the more diverse they are, the more computation it takes (average case) to find a "valid solution". Computation of this sort requires communication, and the higher the bandwidth, the more computation is allowed. Obviously, some people are better or worse at communication, some people are better or worse at computing, and people who know each other well tend to have more efficient communication than people who don't.
Over an internet connection, there is no bandwidth. Text does not communicate quickly enough. Even phone or video chat is a pale ghost compared to actually being there. So, by necessity, there is less of this kind of shared computation. The closest thing to it today is when someone creates something cool and posts it. And that's how games work, too: the developer creates something cool and lets you walk through it.
What they are doing is a mathematical inevitability. They are assembling the chunks to be fewer, larger, and/or less diverse pieces. This allows people to take them in extremely rapidly, with a minimum of communication... and a minimum of computation.
Sure, you can create a fascinating, popular, or mysterious thing which makes people think and laugh and pass it around. But you're not playing with them. You're letting them watch you play.
In other cases, players will still create team content. For example, guilds. They create them inside the constraints they are given, because having a shared reality makes communication more efficient. Even with the extra efficiency, it typically takes weeks or months before a guild really starts to have fun together. The communication is so slow that computation is stretched thin, pieces ooze together over minutes (or, usually, hours) rather than seconds. It works, but it works so slowly.
How can this be fixed? I can see two ways:
Can you make a game with fewer, larger, easy-to-fit-together "story pieces" that players can swap around, get together with, and work together on? For example, instead of having to carefully decide exactly who they are, give them a choice of three pre-defined characters?
Can you make a game where the shared reality is carefully defined to cause the most efficient communication over the limited bandwith you have? A game which either has new methods of communication specifically geared towards putting pieces together, or a game whose reality is designed so that interacting while in that reality essentially causes pieces to fall together - like guilds going on raids and optimizing loot?
Better yet, can you do both simultaneously?
Or did this post not even make any sense?