pcolor() does not show the values as little square pixels like the imaging functions (imshow(), imagesc(), and image()) do.
Basically it's because it's not the center of the little square that has the value, it's the edge of the square that has the value. Then it makes a plane tilted to the other edge with the colors changing along the way. However if you say shading flat, it doesn't change the color, it basically makes the square flat, not tilted, and so the color will be the color of one of the edges (I'm not sure which edge it uses to decide on the color).
So you can see from the above display of a 3x3 matrix, there are 3 edges but only 2 tiles/squares. For this annoying reason, I don't use pcolor() and use imshow() instead. You also might want to consider that if you just simply want to display a matrix. You can still apply a colormap if you want, and one of your own choosing instead of the one pcolor chooses for you, by calling colormap().