You are right. I just benchmarked it. The overhead of the maps far outweighs the cost of the sort. In fact it is so slow I will have to ask p5p about it. Here is my test:

use strict;
use Benchmark;
use vars qw(%hash %test);

for (1..1000) {
    $hash{"key$_"}{this} = "num$_";
}

%test = (
    schwartz => sub {
        my @sorted = map $_->[0],
            sort {$a->[1] cmp $b->[1]}
            map [$_, $hash{$_}{this}], keys %hash; 
    },
    straight => sub {
        my @sorted = sort
            { $hash{$a}{this} <=> $hash{$b}{this} }
            keys %hash;
    },
    stupid => sub {
        my @sorted = sort
            {
                my $foo = [$a, $hash{$a}{this}];
                $hash{$a}{this} <=> $hash{$b}{this}
            } keys %hash;
    },
);

timethese (-1, \%test);
[download]

Believe it or not, stupid is several times faster for me than schwartz. IMO there is simply no possible good reason for that!