If you transcode JPEG into JPEG XL the underlying DCT is limited to the usual 8x8 block (otherwise it'd be much harder to reconstuct the original bitstream), an

If you transcode JPEG into JPEG XL the underlying DCT is limited to the usual 8x8 block (otherwise it'd be much harder to reconstuct the original bitstream), and all the improvement will come solely from better entropy coding and prediction.