Interestingly, using a string is still faster than using a compiled regex (even when I wrap the string in (?:^...) like the compilation does).

#!/usr/bin/perl
use warnings;
use strict;
use feature qw{ say };

my $re = qr/^ *(?:say|print)/;
my $s  = qq/(?^:^ *(?:say|print))/;
open my $in, '-|', qw{ perldoc perlfunc } or die $!;
my @lines = <$in>;
push @lines, @lines for 1 .. 5;
say scalar @lines;

use Benchmark qw{ cmpthese };
my ($ir, $is);
cmpthese(-3, {
    regex  => sub { $ir = 0; /$re/ and ++$ir for @lines },
    string => sub { $is = 0; /$s/  and ++$is for @lines },
});

$ir == $is or die;

__END__
259616
         Rate  regex string
regex  13.5/s     --   -23%
string 17.6/s    31%     --
[download]

Is it because of dereferencing of the regex object?