Survey grade GPS is almost always differential, meaning they place a station at an accurately surveyed point (real time) or use logged data from an error server such as some universities provide. At work, we use a server at IU for correction, which is about 25 miles away. Our stuff isn't surveying, but we need better accuracy than standard GPS. We put in our logged data, and the server goes back and looks at the error it was seeing at the times our readings were recorded and corrects our data.

Back to your original question CP, the shear number of satellites isn't the only important factor. The geometry is important too. Imagine two guys on a football field trying to triangulate a spot. If they are both standing on opposite ends of the 50 yard line and the spot is also on the 50, they will accurately place it on the line, but its exact location on that line is tough. If you place them right next to each other, again they willl have a hard time precisely locating the object. The best scenario puts both guys and the object on different corners of an equilateral triangle. The same situation exists for the satellites, only in 3 dimensions. They work best when they line from each satellite to the receiver is at about 90 degrees to the lines from the other satellites to the receiver. With good geometry, you can get accurate measurements with just a few satellites. More satellites can allow better error correction.

Here's a good link explaining sources of error GPS explained: Error sources