I think it has to do with the interpolation method. Every time you transform a selected area it has to interpolate some pixels to redraw the new image as a result you loose some information. The question is which is the best interpolation method for transforming selections.