The node you've mentioned actually uses an even faster form of the Schwartzian transform I've completely forgot about.

Thus, I've added it to my benchmark too:

sub sort_arr {
        my @h = map -s, @all;
        @results = @all[sort {$h[$a]<=>$h[$b]} 0..@h-1];
}
[download]

The results are amazing: this form is about 1.5 times as efficent than any of the other ones.