Thank you all for the info. It's unfortunate that measurement on
Android is so difficult, given that it's one of the platforms that
stands to benefit the most from power consumption improvements.
(Measurements on desktop are much easier with tools like
tools/power/rapl which access the built-in Intel power estimates.)
The webpage from bug 1200028 might be an interesting test case -- once
the patches from bug 1203766 landed I saw reductions of almost 50% for
processor+memory power on Mac and Linux desktop (though Alice didn't
see any change in CPU utilization on Windows).
Nick