my @sorted = sort @unsorted;
for (my $i=$#sorted; $i>=0; $i--) {
   print $sorted[$i], "\n";
}
[download]

However,
my @sorted = reverse sort @unsorted;
is optimized to be just as fast as
my @sorted = sort @unsorted;
so you're making your program less readable for nothing by avoiding reverse.

use strict;
use warnings;

use Benchmark  qw( cmpthese );
use List::Util qw( shuffle );

use constant COUNT => $ARGV[0];
use constant TIME  => $ARGV[1];

our @unsorted = shuffle map "$_", 1..COUNT;

my $neg_step = '
   use strict;
   use warnings;

   our @unsorted;
   my @output;
   my @sorted = sort @unsorted;
   for (my $i=$#sorted; $i>=0; $i--) {
      push(@output, $sorted[$i]);
   }
   1;
';

my $neg_i = '
   use strict;
   use warnings;

   our @unsorted;
   my @output;
   my @sorted = sort @unsorted;
   for my $i (-$#sorted..0) {
      push(@output, $sorted[$i]);
   }
   1;
';

my $b_cmp_a = '
   use strict;
   use warnings;

   our @unsorted;
   my @output;
   my @sorted = sort { $b cmp $a } @unsorted;
   for my $i (0..$#sorted) {
      push(@output, $sorted[$i]);
   }
   1;
';

my $reversed = '
   use strict;
   use warnings;

   our @unsorted;
   my @output;
   my @sorted = reverse sort @unsorted;
   for my $i (0..$#sorted) {
      push(@output, $sorted[$i]);
   }
   1;
';

cmpthese(TIME, {
   neg_step => $neg_step,
   neg_i    => $neg_i,
   b_cmp_a  => $b_cmp_a,
   reversed => $reversed,
});
[download]

outputs

>perl 589197.pl 1000 -3
          Rate neg_step  b_cmp_a reversed    neg_i
neg_step 301/s       --      -0%      -1%      -1%
b_cmp_a  302/s       0%       --      -1%      -1%
reversed 304/s       1%       1%       --      -0%
neg_i    304/s       1%       1%       0%       --

>perl 589197.pl 1000 -3
          Rate  b_cmp_a neg_step reversed    neg_i
b_cmp_a  301/s       --      -1%      -2%      -2%
neg_step 303/s       1%       --      -1%      -1%
reversed 306/s       2%       1%       --      -0%
neg_i    306/s       2%       1%       0%       --

>perl 589197.pl 1000 -3
          Rate neg_step  b_cmp_a    neg_i reversed
neg_step 301/s       --      -0%      -1%      -2%
b_cmp_a  302/s       0%       --      -1%      -1%
neg_i    305/s       1%       1%       --      -0%
reversed 306/s       2%       1%       0%       --

>perl 589197.pl 10000 -5
           Rate  b_cmp_a neg_step    neg_i reversed
b_cmp_a  22.4/s       --      -1%      -2%      -3%
neg_step 22.6/s       1%       --      -2%      -2%
neg_i    23.0/s       2%       2%       --      -0%
reversed 23.0/s       3%       2%       0%       --

>perl 589197.pl 10000 -5
           Rate neg_step  b_cmp_a reversed    neg_i
neg_step 22.5/s       --      -0%      -2%      -2%
b_cmp_a  22.5/s       0%       --      -2%      -2%
reversed 22.9/s       2%       2%       --      -0%
neg_i    23.0/s       2%       2%       0%       --

>perl 589197.pl 10000 -5
           Rate  b_cmp_a neg_step    neg_i reversed
b_cmp_a  22.5/s       --      -1%      -2%      -2%
neg_step 22.7/s       1%       --      -1%      -1%
neg_i    22.8/s       2%       1%       --      -1%
reversed 23.0/s       2%       1%       1%       --
[download]

As you can see, none is faster than any other, so pick the one that's more readable and maintainable.

Update: Oops! Changed $i++ to $i--.
Update: Added benchmarks.