ikegami, I know you already know it, but reading your post somebody could thing that using the your GRT code is a good idea because it is the faster solution when actually it is not!

It is not good because using substr() to extract the sorting keys is very weak. If for instance, the file names extensions are changed, it will extract incorrect keys and, what is worst, without reporting any warning or error to the user!

Even if slower, using a regular expression to extract the keys is probably the best solution.

In any case, there is a faster (not recommendable either) method...

...
use Sort::Key qw(keysort);
...
cmpthese(-3, {
    ...
    sk => '
        use strict; use warnings;
        my @sorted = keysort { substr($_, -(19+20), 19) } @::LogFileLi
+st;
    ',
});
[download]

that on my computer runs as...

            Rate memoized       st      grt    naive       sk
memoized  8784/s       --     -26%     -51%     -54%     -58%
st       11894/s      35%       --     -33%     -38%     -43%
grt      17794/s     103%      50%       --      -7%     -14%
naive    19207/s     119%      61%       8%       --      -7%
sk       20714/s     136%      74%      16%       8%       --
[download]

Note also that on my hardware, naive is actually faster than grt.