I think that there are two very different considerations that are engaged when considering the worthiness of a setting.
Firstly, you have the consideration of the richness and consistency of the setting itself. In a very real sense, the "setting" can be a character in the tale. It has a complex and detailed backstory, and the differences and flavors that the setting imparts to the tale should be almost tangible.
The best examples of these sorts of settings are drawn from fantasy literature and the great "world builders": Tolkien's Middle Earth, Donaldson's The Land, Martin's Westeros, etc.. Each is a detailed and rich setting which brings so much to the table that it is almost a character in the tale itself. I think that many homebrew GMs prefer this sort of verisimilitude that these settings convey to the reader and they aim to reach that level of flavor for their own campaign worlds. Some of them look for the same aspects in published game worlds, too.
But there is a second and, I would argue, overriding design aspect to published game settings: they need to work as a setting for all manners of divergent role-playing game sessions. You would think that this should be glaringly obvious but it seems pretty clear from the posts in this thread so far that... well... maybe it isn't so obvious after all. Because the sorts of rich and distinctive detail that fantasy authors bring to their worlds within their own fiction are so rich -- and so distinctive -- that they don't work for the large majority of RPGs.
Some gamers like high fantasy dungeon crawls. Others like urban fantasy tales as might be inspired by Fritz Leiber's city of Newhon where Fafhrd and the Gray Mouser's adventures took place. Some prefer a more modernized urban fantsay a la China Mieville. Some others prefer a story where court intrigue is the main theme. Others still prefer their tale to be about challenging authority and resisting the Evil Overlord. The list goes on and on and on and the Devil is in those Details.
On and on it goes. There are DOZENS AND DOZENS of aspects to game setting and themes which "best match" the preferred style of a given DM/GM and their players. One size does not and cannot ever fit all. It cannot be done. EVER.
So, recognizing the fact that different styles of games requires different settings, the most successful published game worlds approach the question of setting by using a kitchen sink approach. The game world is NOT a game world, rather, it's 20 or more "game worlds" forced on to one planet with a label and branding thrown on top of it all. These regions each feature different cultural aspects to the setting, but their real purpose is to permit vastly different campaign styles and themes to all be set within the same "world". As long as you don't try and make too many of these regions and settings interact with each other TOO much -- it all mostly works.
Is it realistic? No, it isn't and it's not even close. But the kitchen-sink approach is the best commercial context to present to tens of thousands of GMs, each of whom have very different campaign goals and styles. In order to do that, you need details -- but not too much detail. You need a story or metaplot -- but not too much story or metaplot which crowds out and overwhelms the story the GM and the players want to tell in their own games.
When you add official fiction set within that game world into that product mix, inevitably, it adds detail and richness -- while at the same time that detail and metaplot crowds out and forestalls other possibilities. You cannot have the good aspect of that detail without adding in the harmful aspect of that detail, too.
We've seen all kinds of these examples in the past. Krynn was created to tell a very specific story for the Dragonlance novels. The more that was written about it, the less tractable the setting became until metaplot overwhelmed the capacity of the setting to permit GMs to tell their stories there.
The Forgotten Realms suffered and ultimately collapsed under the same weight of fiction and modules set within it. While Bob Savatore's novels sold extremely well over time, that success began to taint and overwhelm the flavor in the whole of the game world, too. The Forgotten Realms, in many ways, became a victim of its own success. I would argue that all successful game worlds experience this same problem.
Paizo's Golarion, which is another incarnation of the Forgotten Realms "multi-region, multi-kingdom design" (the Kitchen Sink Approach) to a world setting has not collapsed under the weight of the burgeoning product line (yet). Golarion may be only five years old, but there have been a staggering number of modules written about the world already. The problem of "which Adventure Path(s) have already happened" question is starting to make that structure groan and shift under its own weight, too. So far, it's still working well. So far, that is.
It seems to me that the single greatest element that can "destroy" the usability of a game world is a very successful line of novels. Once some specific novels become widely read by players and GMs alike, the flavor and canon in that published tale begins to overwhelm the capacity of the game world to be a "blank enough slate" for PCs to be the heroes at the table. Instead, the role of the REAL "hero at the table" is a character in a series of novels that have already been written and in which we all know how the story ends, too. That's BAD for adventure gaming.
All by way of saying, there is no "best" setting because the nature of what makes a setting attractive is highly dependent upon the type of game you want to run -- and we don't necessarily want to run the same games. Even when we do share many of the same objects and goals for our campaigns, our groups may have very different play styles, too.
I will, however, say this: if you want to run one of Paizo's Adventure Paths, the world of Golarion is designed to work extremely well with those APs. Not only that, but it actually works, too.
So far, that is. Add one extremely successful series of novels into that mix and all bets are off.
