OK, let's examine your example. On a 2 x 2 x 2 cube, the coordinates for the hole would be 1.0000 from one side, and 1.0000 from adjacent side. The old way was to either bind you with the title block +/- 0.010 (or now 0.005) or a non-title block tolerance of +/0.002 on both axis. This gives you a placement zone that is square that limited to a deviation of a maximum of 0/.002 error in either axis. But since you'll see this type of tolerancing many times on round parts, and holes where things bolt together, it doesn't make sense to have a Cartesian tolerance system. If you drew a circle, centered on the true position of the hole (no errors considered) at the radius that included the corners of that theoretical square, you'll see it has more area than the square. Now consider a circle inside that square. Now you have a smaller zone, right? The are in the corners is not included now. On the larger circle, you can see that as you move away from the centerline of one of the axis, you have to adjust your position to follow the arc and stay within the circle. But it allows you to be further out on one axis that a Cartesian method, provided you are closer on the other axis. As long as both parts use the same tolerancing system and tolerance zones, you actually gain room for error and still have a workable part.
Clear as mud now, eh? I'll walk to the shop in a bit and see if what I have is worth scanning.