If you accept the principle that code in modules is destined to be reused in lots of situations, including performance critical ones, and should therefore do it's utmost to be as efficient as the author is comfortable maintaining, then List::Util's use of aliasing to sqeeze a little more performance seems entirely reasonable.

That said, they missed a trick. Using a block form map creates a additional level of scope that they aren't using and slows performance.

They can squeeze another 30%, and almost 50% over the 'naive' implementation, by avoiding it:

sub buk (@) {
    my @a = \( @_ );
    my $n;
    my $i = @_;
    map+( $n=rand($i--), ${ $a[ $n ] }, $a[ $n ]=$a[ $i ] )[ 1 ], @_;
}

cmpthese -1, { map { $_ => "$_ 1..1000" } qw/naive listutil buk/ };

__END__
C:\test>junk
          Rate    naive listutil      buk
naive    520/s       --     -13%     -32%
listutil 597/s      15%       --     -22%
buk      769/s      48%      29%       --
[download]