I have one more optimization for you MeowChow. Inside your _min() and _max() subroutines, if the test returns false, it reassigns the original value back to itself. You can get around a 25% speed increase in the algorithm if you only assign the $_ value to $max or $min if it passes the test.

Here's a short benchmark that illustrates the difference better:

#!/usr/bin/perl -w

use Algorithm::Numerical::Shuffle qw(shuffle);
use Benchmark qw(cmpthese);

#Randomize the array contents
my @array = shuffle 0..1000;

cmpthese(-3, {
  max_dkubb1 => sub { max_dkubb1(@array) },
  max_tilly  => sub { max_tilly(@array)  },
});

sub max_dkubb1 {
  my $max = shift;
  $max < $_ and $max = $_ for @_;
  return $max;
}

sub max_tilly {
  my $max = shift;
  $max = $max < $_ ? $_ : $max for @_;
  return $max;
}
[download]

On my system the max_dkubb1() routine runs just over 25% faster than the other.