Comment by ska

The opposite is likely. RLE works best with long "runs" of the same value. Dithering will tend to break up these regions if it improves quantization error.