Hmmm... I'm afraid that the recursive 'between' above there might not work for complex cases. The non-greedy regexp would make it match the first start-end pair it found, so if we had:

yadda yadda start this is comment start this is still comment end this
+ should still be comment end yadda yadda
[download]

then we should wind up with the whole thing, minus start and end and yadda, but instead we get:

 this is comment start this is still comment
[download]

The only way I can think of to get around this is by keeping external track of the levels. This also de-recurses it, which makes it less beautiful, but faster (in theory):

sub between {
    my ($text) = @_;
    my $level = 0;
    my @comments = ();

    while ( $text =~ m{\G .*? (start|end) (.*?) (?: (?=start|end) | $ 
+) }gxs ) {
        if ( $1 eq 'start') { $level++; }
        else { ($level > 0) and $level--; }

        $level > 0 and push(@comments, $2);
    }

    return join('', @comments);
}
[download]

This returns:

this is comment  this is still comment  this should still be comment
[download]

So what we're doing here is going through the text looking for 'start's and 'end's. We keep a counter indicating how many levels deep we are in 'start's and 'end's. Every time we hit a 'start', we add one. Every time we hit an 'end', we subtract one, checking first to make sure that our level doesn't go negative. (Otherwise, somebody could mess us up by starting a file "end end end".)

Afterwards, we look at the patch of text between the current tag and the next start/end tag. If our level is greater than 0, we're between a 'start' and an 'end' tag, so we store that segment. Otherwise, we're not, so we look for another 'start' or 'end' tag until the end of file.