The following benchmark might be interesting:

#!/usr/bin/perl

use strict;
use warnings;

use Fcntl qw /:seek/;
use Benchmark qw /timethese cmpthese/;

my $lines = shift || 50_000;
my $file  = "/tmp/big";

# First create a big data file.
open my $fh => "> $file" or die "open: $!";
foreach (1 .. $lines) {
    foreach my $i (1 .. 3) {
        my $r = int rand 3;
        print $fh "abcdef" if $r == 0;
        print $fh "123456" if $r == 1;
        print $fh "<[{}]>" if $r == 2;
    }
    print $fh "\n";
}
close $fh or die "close: $!";

my $words   = "/tmp/words";
my $digits  = "/tmp/digits";
my $punct   = "/tmp/punct";

my @array   = ([$words  => '[a-z]'],
               [$digits => '[0-9]'],
               [$punct  => '[^a-z0-9]']);


sub many {
    foreach my $entry (@array) {
        open my $fh => ">", $entry -> [0] . ".m" or die "open: $!";
        close $fh or die "close: $!"; 
    }
    open my $fh => $file or die "open: $!";
     
    while (<$fh>) {
        foreach my $entry (@array) {
            my ($f, $r) = @$entry;
            if (/$r/) {
                open  my $fh1 => ">> $f.m" or die "open: $!";
                print $fh1 $_;
                close $fh1 or die "close: $!";
            }
        }
    }
}

sub one {
    open my $fh => $file or die "open: $!";
    
    foreach my $entry (@array) {
        my ($f, $r) = @$entry;
        open my $fh1 => "> $f.o" or die "open: $!";
        seek $fh => 0, SEEK_SET or die "seek: $!";
        while (<$fh>) {
            print $fh1 $_ if /$r/
        }
        close $fh1 or die "close: $!";
    }
}
                
                
cmpthese -60 => {
    one      => \&one,
    many     => \&many,
};

unlink $file or warn $!;
unlink map {my $s = $_ -> [0]; ("$s.m", "$s.o")} @array;

__END__
     s/iter  many   one
many   5.74    --  -95%
one   0.267 2045%    --
[download]

Going through the file repeatedly, and printing out the files one by one is a big winner.

Abigail