Here's some trivial benchmarks that at least compare comparable operations and ones that are just slightly more than trivial:

#!/usr/bin/perl -w
use strict;
use Benchmark 'cmpthese';
my @a = 1..1000;
sub p {
    my( $a ) = @_;
    if( $a & 1 ) {
        return $a*$a;
    } else {
        return ($a-1)*($a+1);
    }
}
cmpthese( -2, {
    for  => sub { my $t = 0; for( @a ) { $t += p($_) } return $t },
    map  => sub { my $t = 0; map $t += p($_), @a; return $t },
} );
cmpthese( -2, {
    map  => sub { my @b = map p($_), @a; return $b[-1] },
    for  => sub { my @b; for( @a ) { push @b, p($_) } return $b[-1] },
} );
[download]

And here are my results:

      Rate for map
for 2101/s  -- -1%
map 2127/s  1%  --

      Rate map for
map 1981/s  -- -0%
for 1987/s  0%  --
[download]

So I don't even have to give my explanation of why Benchmark.pm saying "20% faster" almost never has any practical meaning.