Your first variation seem to destroy the array that is handed to it. Dump the original array before and after your reverse and see. Fast, but kinda rude. =)

Your second variation seems to do it in line, reversing the original array! I hadn't thought about making one that does it all in place! =)

Your third variant was the one that left the original undisturbed.

I really like the map variation, and I like better never making a named variable:

use Data::Dumper;
my @names = (1, 2, 3, [4, 5, 6, [7, 8, 9], 10], 11, 12, [13, 14]  );
print Dumper(@names);
my $reved = rev_arr5(\@names);
print Dumper(@names);
print Dumper(@$reved);

# less than 80 col! yay!
sub rev_arr5 {
      return [map { ref eq 'ARRAY' ? rev_arr5($_) : $_ } reverse @{shi
+ft()}];
}
sub rev_arr2a {
      @{$_[0]} = map { ref eq 'ARRAY' ? rev_arr2a($_) : $_ } reverse @
+{$_[0]};
}

sub rev_arr4 {
        my $arr = [reverse @{shift()}];
        foreach (@$arr) {
                if ( ref $_ eq 'ARRAY' ) {
                        $_ = rev_arr4 ($_);
                }
        }
        return $arr;
}
[download]

That version is safe from unintended effects (plays well with others) and is lickety-split. BTW I ran 1 million reps of eash thru Benchmark::cmpthese so really this is a waste to optimize anyway. =) I also tested a variant of Ted's number 2 that has no return at the end, since it modifies in place anyway.

                   Rate Extremely_orig Extremely_4 Ted_var_3 Extremely_5 Ted_var_2 Ted_var_2a
Extremely_orig 106045/s             --         -3%      -28%        -46%      -47%       -51%
Extremely_4    109170/s             3%          --      -26%        -45%      -45%       -50%
Ted_var_3      147059/s            39%         35%        --        -25%      -26%       -32%
Extremely_5    196850/s            86%         80%       34%          --       -1%        -9%
Ted_var_2      199601/s            88%         83%       36%          1%        --        -8%
Ted_var_2a     217391/s           105%         99%       48%         10%        9%         --

My/Ted's one liner beats all the other variations that protect the original array quite handily. Ted's cool in place matches the one liner(ish) and removing the return nets you a small but not insignificant speedup. My variation 4 just shows the gain from one less assignment.

Fun Stuff! Doh! last minute change, this variant:

# Extremely_5a   209644/s
sub rev_arr5a {
      return [map { ref eq 'ARRAY' ? rev_arr5a($_) : $_ } reverse @{$_
+[0]}];
}
[download]

exploits the "no shift" trick to beat even the regular in place Ted_var_2 by 4%, Extremely_5 by 5%, and comes in about 6% under the in place Ted_var_2a.

Benchmark is sucking my will to live. I'm gonna be here all night if I don't stop!

--
$you = new YOU;
honk() if $you->love(perl)