Pixel shift is supposed to increase quality on the pixel level, from such low-res images most people couldn't tell the difference between a K3 and a 645Z. The file-size is probably smaller because it has less noise. Noise is random so you have to store each individual pixel separately, while you can use algorithms for repeating patterns (such as a wall). Just like ISO 800 files are bigger than ISO 100 files are. Dunno where those extra 48 million subpixels went though... As far as I know Bayer DNG's only store one color per pixel and the RAW converter has to demosaic the picture, so RGB DNG should be larger in size. Feel free to correct me if I said something silly
Last edited by Solaire; 05-24-2015 at 10:59 AM.