Just for fun, here's the benchmark. It took some guesswork to get it to run the subs in the right order - clean first, then use English;, then naughty. If anyone uncomments the print statements to test the order, use 1 as an argument so you don't have to wait forever. Here are the results:

use strict;
use Benchmark qw/cmpthese/;

my $time = shift || -5;
my $text = 'x' x 10_000;
sub clean {
#    print "clean";
    $text =~ m/^x/;
}
sub make_dirty {
#    print "md";
    eval "use English;";
}
sub naughty {
#    print "naughty";
    $text =~ m/^x/;
}
my %hash = ( clean => 'clean',
             naughtify => 'make_dirty',
             sawamp => 'naughty',
           );
cmpthese ( $time, { clean => 'clean',
                    naughtify => 'make_dirty',
                    sawamp => 'naughty',
                  });
__END__
results:
C:\s\pldir>naughty.pl -5
              Rate naughtify    sawamp     clean
naughtify    433/s        --      -98%     -100%
sawamp     24153/s     5481%        --      -92%
clean     300603/s    69366%     1145%        --
[download]

Someone with more benchmark-fu may correct me on this, but it looks right to me.