Adding to what jack says, each sensor read out has to be compressed into a frame at 1/30th of second or more, to do this if you compress the data to 8 bits in camera it often simply chops the extra bit depth off with a generic Pixel combination algorithm that results in a simply dumps half the data down the toilet. It's basically cutting the resolution of color data in half. You may have 1920 pixels but 970 of them are just copied from another pixel to make it lower bit depth.
In post instead, the software only takes away data where pixels are the same already to conserve image quality. It's two totally different results.
