G'day fredo2906,

I created some test input, before seeing your updated OP, as follows:

$ cat > pm_1075438_1.txt
qw<er ty>
a>sd <fgh

$ gzip pm_1075438_1.txt

$ cat > pm_1075438_2.txt
<zxc vbn>
123> <456

$ gzip pm_1075438_2.txt
[download]

This script removes the need for the intermediary @o and @l:

#!/usr/bin/env perl

use strict;
use warnings;

my %h;
++$h{$_} for map { chomp; $_ } `zcat @{[glob '*.gz']} | sed 's/[<> ]//
+g'`;

use Data::Dump;
dd \%h;
[download]

Output:

{ 123456 => 1, asdfgh => 1, qwerty => 1, zxcvbn => 1 }
[download]

FWIW, this line:

++$h{$_} for map { chomp; s/[<> ]//g; $_ } `zcat @{[glob '*.gz']}`;
[download]

produces identical output. I'll leave you to benchmark (if you want).