"without using HTML::TokenParser"

Why? This is simply not the right decision. In this case, it is more important to do it right, with the right tool - HTML parser (for example what murugu mentioned), but not strugling with the "right regexp".