Could you provide some input HTML that's valid, and explain by what rules the <h3> got wrapped up in a <div> along with the two <p>'s, instead of the <p> preceding the <h3> being wrapped along with the preceding <h2>?

Anyway, I agree that Mojo::DOM is nice. Maybe this is something to get started with:

use warnings;
use strict;
use Mojo::DOM;

my $html = "<h1>a</h1>\n<p>b</p>\n<h2>c</h2>\n<p>d</p>\n"
    ."<h3>e</h3>\n<p>f</p>\n<h2>g</h2>\n<p>h</p>\n<h2>i</h2>\n"
    ."<h1>j</h1>\n<h2>k</h2>\n";
my $dom = Mojo::DOM->new($html);
$dom->find('h2')->each(sub {
    my $next = $_->next;
    my $new = $_->wrap('<div></div>');
    if ( $next && !$next->matches('h1') ) {
        $new->append($next);
        $next->remove;
    }
});
print "$dom";
[download]