You're right

use strict;
use warnings;

use Benchmark ();
[download]

sub test_arr  { my @arr = @_;       $a = join('|', @arr); }
sub test_ref  { my @arr = @{$_[0]}; $a = join('|', @arr); }
sub test_ref2 {                     $a = join('|', @_);   }
[download]

{
   my @tiny = qw( abc def ghi );

   Benchmark::cmpthese(-3, {
      arr  => sub { test_arr   @tiny },
      ref  => sub { test_ref  \@tiny },
      ref2 => sub { test_ref2 \@tiny },
   });
}

{
   my @small = ('abcdefghi') x 10;

   Benchmark::cmpthese(-3, {
      arr  => sub { test_arr   @small },
      ref  => sub { test_ref  \@small },
      ref2 => sub { test_ref2 \@small },
   });
}

{
   my @giant = ('abcdefghi') x 1000;

   Benchmark::cmpthese(-3, {
      arr  => sub { test_arr   @giant },
      ref  => sub { test_ref  \@giant },
      ref2 => sub { test_ref2 \@giant },
   });
}


__END__
output
======
[download]

TINY ARRAY

         Rate    ref    arr   ref2
ref  120870/s     --   -14%   -17%
arr  140004/s    16%     --    -4%
ref2 145146/s    20%     4%     --


SMALL ARRAY

         Rate    ref    arr   ref2
ref   61587/s     --    -8%   -58%
arr   67214/s     9%     --   -54%
ref2 146895/s   139%   119%     --


GIANT ARRAY

         Rate    ref    arr   ref2
ref     850/s     --    -0%   -99%
arr     850/s     0%     --   -99%
ref2 148539/s 17383% 17381%     --
[download]

The arrays would have to be pretty damn big for it to matter if one does my @arr = @_.