I wouldn't use search and replace to detect the stop word, it's doing much more work than you need and is going to get really slow as your stop word list increases.

If you store your words in a hash then it's easy and efficient to test if a word exists in the hash, so then you can do something like this

-- err not tested :)

my @words = qw/ and or not one two three/;
my %stop;
$stop{$_}++ for @words;

...

for my $w (split /\s+/, $line) {
    next unless length($w) > 2;
    next if $stop{$w};
    ...
    $keys{$w}++;
}
[download]

update -- fix typos