Every browser that handles HTML 4 supports a kind of "tag soup". You must take explicit action to shift them into a strict mode; consequently, it does not take longer to "parse a broken schema" -- handling an HTML document without the final closing tags should be no slower than a complete document, because no validation is taking place.
HTML is not XML.
It might be invalid, it is "monstrously invalid XHTML", but it is not "monstrously" invalid HTML.
I think I've spent so many years battling with anti-standards folks that I now take it personally when people advocate leaving out tags and not using a proper Doctypes and the countless other things Google does. In this case, I failed to do any research before making my argument, and have been deftly slain for it ;)
I'd like to see Google provide facts and figures. When Steve Souders worked at Yahoo! a large part of recommendations were backed up by experimental data collected by Tenni Theurer (my manager's wife, coincidentally).
I'd like to see Google provide the same kind of experimental data for their guidelines. If saving 14 bytes is a network saving, where are the page render tests in browsers, particularly "A-grade" browsers?
In light of the whipping their "facts" on PHP optimization recommendations have received here ( http://news.ycombinator.com/item?id=676856 ), it's best not to take pronouncements from a google post as coming from on high.
Well, the question I answered was whether it was valid. I'm not knowledgeable enough to answer the speed question.