When I was in physics, professors would always tell us to do a back of the napkin common sense estimate before undertaking any actual calculations. Eg, what would we expect the ballpark to be to achieve clinically acceptable fits. Here is my "back of the napkin" calculation to figure out what an acceptable theoretical "limit" for scanner accuracy and data density should be.
First of all accuracy...we can all agree that we would hope for infinite accuracy or zero uncertainty in our measurements. Granted, for us there is a practical limit at which point we have approximately perfect accuracy and in my opinion, that limit is 1 micron since that is on the order of the diameter of a dentine tubule.
As scott, disturbed and K2 have stated, more data density yields more detail. But EWH and some others have stated that there is a point of diminishing return. Eg, if we are more precise than our manufacuring process...what's the point? But lets say that the manufacturing process is perfect....how many points do we really need? My answer is, as many as needed to resolve the smallest detail we are interested in. Eg, sharp margins, little dips and chips etc. Here is how I would think of it. A new explorer tip diameter is around 100 microns, I would like to be able to scan something half that diameter (eg, the tip will not fall into a hole that size and thereby be undetectable). So, to resolve a 50 micron sphere (or a 50 micron circle in 2d to be simpler),how many points do we need? How about 8 (eg, a stop-sign represents a circle fairly well?). If we space 8 points around the circumfrence of a 50 micron circle, they will each be (50*pi/8) about 20 microns apart. So to me, 20 micron spacing is probably a good limit to aproach if a new explorer is your barometer....but you only need that density at your sharpest features. So, do not be worried if your stl's have wide point spacing on relatively flat surfaces as your software may have smartly thrown them away realizing they are like two state employees doing the same job.
Furthermore, this is how CEREC gets acceptable fits for some docs. If you look at the preparation guidelines, a summary is...."prep so that nothing is sharp and the margins are flat." If there are no sharp curvatures, you can represent an object with sparse point spacing. Scott said before, "what's going on in that 100-200 micron gap between points." And in some cases, the answer is "nothing because my preparation is very smooth and ideal." Other times it's "a whole lot, expect that restoration to bind or be open here."