What potential overhead? Perl will *not* recompile the regex in each iteration (unless you modify $from or $len). Here's a proof, assuming the program is in the file /tmp/qq:

    $ perl -Dr /tmp/qq 2>&1 | grep Compiling
    Compiling REx `^(.{2}).{2}(.*)$'
    $
[download]

It only gets compiled once.

Here's a benchmark:

#!/usr/bin/perl
                     
use strict;
use warnings;
                     
use Benchmark 'cmpthese';
    
our @data = <DATA>;
chomp @data;
                    
our $from = 2;
our $len  = 2;

our (@a, @b, @c);

cmpthese -10 => {
    substr   => 'for (@a = @data) {
                     substr $_, $from, $len, "" if length ($_) >= $fro
+m + $len
                 }',
    regex    => 'for (@b = @data) {
                     $_ = $1 . $2 if /^(.{$from}).{$len}(.*)$/;
                 }',
    regex_qr => 'my $qr = qr /^(.{$from}).{$len}(.*)$/;
                 for (@c = @data) {
                     $_ = $1 . $2 if /$qr/;
                 }',
};  

die "Unequal" unless "@a" eq "@b" && "@b" eq "@c";
                    
__DATA__
12345678
yellow
a
xx
yyy
a somewhat longer line to compensate for the smaller lines.
bla bla bla bla



            Rate regex_qr    regex   substr
regex_qr 38207/s       --      -3%     -62%
regex    39497/s       3%       --     -60%
substr   99555/s     161%     152%       --
[download]

Abigail