DiapDealer,
You have probably already considered this, but some potential issues that popped into my head.
What about conversational exchanges between characters like:
<p>"Yada, yada yada yada," he said.</p>
<p>"Blah! Blah, blah blah, blah blah blah</p>
<p>"Yada yada."</p>
While I've used paragraphs in this example they could be spans or divs. But wouldn't the non-value add stripping run the risk of merging this conversation into a single run-on comment by the first speaker?
Also the less frequently seen monologue of multiple paragraphs, which lacks a closing quotation mark until the final paragraph.
(I haven't downloaded the PI as yet - I've been busy at work - so perhaps these points are already accounted for.)