Early in Mark Wolf's awesome new book about subcreation, he makes note of Eco's discussion of cult products. Wolf highlights the fact that a good cult film is not necessarily completely coherent. Rather, it has chunks and pieces that allow the audience to participate in creation. A good world is not a clean story from end to end, its a lattice with hooks for people to hang other things. As we all know from building toys, anything you hang on a hook should itself be a hook.
A few weeks ago, we discussed what makes good computer game worlds different from the current over-designed worlds we are getting. It's not necessarily free-form play or anything, but rather the presence or absence of features that provide hooks for our own immersion. A huge sandy desert is a great sandbox but, pace ATITD, is not a great world. A game filled with narratives and achievement ladders is also not a great world. A game with a million loose ends is not a great world. A game that has ends that are not loose but rather awaiting further development or discovery is the ticket. If there are mountains at the edge of the place space, we should not be told that over the mountains is "a place nobody goes." We should be told that "Across the mountains lie the sands of Khalibar; people who go there never come back. We do not know what happens to them." That's why it is so disheartening to open a new game and pop up the map, and see the entire world including a clearly defined boundary, beyond which is nothing. Once I see my progress on that map, I can't help extrapolate and sense how long it takes to explore the whole thing. It's kind of like, for an achievement player to kill the first mob and, from that experience, get a good guess as to how the final boss fight will go. Deflates the mystery. So: Get rid of the boundaries!
Anyways, you shoujld pick up and read Building Imaginary Worlds. You'd be surprised how many cool worlds there have been prior to Ultima.