Theoretically, to get a more precise and sharper image registration it is better to have possibly smaller pixels, or else each pixel registers image area instead of point. Because theoretically if you divide image in pixels, the exact , single pixel information about image is projected at one point.. area is some sort of aproximation or integration over photo detector surface, which isnt exact value of the point correspongin to the pixel.
Practically you face many problems when decreasing pixel size (like noise), so in reality no one makes tiny pixels on large sensor, but they get smallet with increased resolution.
Btw, did you know that in a full frame CCD, only 1/3 of the surface is used for pixels. Even less in other ccd types.
This is another reason, why one would like to use more pixels than needed:
http://en.wikipedia.org/wiki/Oversampling