There has been some discussion on the blogs recently regarding the value and practicality of preparation. Check out Roger, in The Mediocrity of Improvisation; Noisms, in Verfremdungseffekt; Courtney, in On the Preparation; and the RPG Site thread which prompted the Monsters & Manuals post. Roger makes the most direct claim for the value of preparation (which I think is indisputable):
In short, while improvised content can be wildly fun and creative, it usually also tends toward a middle ground of risk and reward. … As sole authority, there is a strong pull toward the middle ground – to mitigate challenges, to clip rewards. The lurking spectre in the background is that of the juvenile, “mad god” style of DMing, where party-killing traps and mind-numbing treasures are handed out, “just because.” Avoiding this spectre, you veer towards the safe and average.
If you are deciding what PCs encounter “just in time” (without rolling on a table, as populating an encounter table is a form of preparation), there is almost no way to be impartial. Your choice of kobold or dragon is largely a choice about whether you want the party to live or die. All of the linked articles above make valuable points, but don’t mention (at least explicitly) what I consider to be the main reason that preparation is necessary, and the main benefit to be gained thereby.
Much of the discussion around sandbox play revolves around the idea of making player choice meaningful (as opposed to railroading where you will tell your story no matter what the players decide, or using quantum ogres). For this kind of play to be fair, PCs must have a way to learn about the threat levels of their possible choices. Players cannot make an informed and meaningful choice if you decide what is present after they have already made choices.
Or, in more concrete terms, you can put Mordor on the map, which gives players a chance to learn about the danger through rumors or other clues, or you can improvise them walking into Mordor. Following Roger, this will either be bland (because you don’t want to kill them) or unfair (because they were not given a chance to assess the risk and prepare, or go somewhere else).
This very specific kind of virtue is not dependent upon the quantity of preparation. You can put in a lot of time into Tolkien-style world building and still not obtain the primary virtue available from preparation, which is threat level communication. For example, a very simple hex map along with four encounter tables (one for each compas direction) and a few lairs are enough for you to be able to impartially dole out info about risk levels if your players are curious.
If you put a little more work into it (maybe some history and upcoming events, to bring the temporal dimension in), your players can make even more sophisticated choices (do they want to attempt the mountain pass before the snows, or take this other lucrative job first?). If PCs still kick in doors without listening first (or take analogous actions in other environments), then what happens is on them, as long as the info was available for them to find in the first place.