Thanks for your feedback. (Author here.)

By my calculations (which I'll machine-check now with Maxima), the 30% and 70% speed-up claims are sound. First, let's fire up Maxima:

    $ maxima
    Maxima 5.30.0 http://maxima.sourceforge.net
    using Lisp SBCL 1.1.8-2.fc19
    Distributed under the GNU Public License. See the file COPYING.
    Dedicated to the memory of William Schelter.
    The function bug_report() provides bug reporting information.

Now let's calculate how much faster we made the row-copy code by unrolling its loop:

    (%i1) orig_loop_speed:     (1*row) / (157*cycle)$
    (%i2) unrolled_loop_speed: (1*row) / (120*cycle)$
    (%i3) unrolling_speedup:   unrolled_loop_speed / orig_loop_speed,  numer;
    (%o3)                          1.308333333333333

In other words, the unrolled-loop speed is 1.3 times the original-loop speed. That's 30% faster, right?

Likewise, how much faster did shaving those 1200 cycles make the tile-copy code?

    (%i4) copy2_speed:        (1*tile) / (2893*cycle)$
    (%i5) copy3_speed:        (1*tile) / (1689*cycle)$
    (%i6) copy3_vs_2_speedup: copy3_speed / copy2_speed,  numer;
    (%o6)                          1.712847838957963

I'm willing to believe that I've screwed up here, but I can't see where. Can you show me?

Thanks for your help!