F1 - Image . i too think is the way to go.

also you could check up on a demo called: "walk_back" something, dont remember the exact name, if you havent figured out the walking part of it.

Before I talk about the concept itself, let me address your specific example; in that you'll probably want 2 NPCs, each at the different location, and toggle visibility of them when a player advances the quest, no? Since if you move the original NPC and make it invisible to players, how will they take on that quest as well?

Images are indeed pretty much the only way to go, but don't expect the concept to be fully functional or easy to fully do properly. Images only handle the visual part. The results of using images like this (to make an atom visible only to certain players) are as follows:
-No player sees the normal NPC's atom appearance. This means its icon, overlays, etc are not visible (because you'll have to null them out, or else everyone will see the object).
-Certain players you choose will see a static graphic that is pasted on (a little like an overlay) on the object. The image itself. Those that you don't let see the image won't be aware of the object visually at all, because otherwise it's invisible (icon-less).
Also, take note nothing here still prevents players that as a result of your 'trickery' can't see the players from being able to interact with it. You have to account for that throughout your code, in all instances where a player can interact with an object, and remove players' abilities to interact with it. This is not always possible, though - say you have a good-ol' Talk() verb on the "invisible" NPC. It's set to automatically be available when a player is near the NPC. Now, even if you change the actual verb to not do anything if the player using it shouldn't see the NPC, the player will still have the verb become available (and most likely 'pop up') whenever he's near the NPC, and there's not much you can do about that other than just not using such verbs.

Note invisibility is more suitable for this than images and can also solve some problems like the no-using-stuff-like-overlays drawback and the above verb problems. But the problem is that invisibility isn't useful in this manner if you use it for a lot of objects, because in short, basically your control over what is invisible and what isn't is quite limited (I wish they'd change this, really). Unless all of the objects you want such 'selective visibility' for appear in a linear sequential matter, you couldn't really pull it off.