I'm not sure, CSS selectors are already matched left-to-right so you're not scanning more stuff, it's just that you're selecting an other level of your match tree.

    ul > li > p

    ul > $li > p

have the same matching complexity: the browser finds all p on the page, then checks if their parent is a li, then checks if that parent is a ul. The difference (which may or may not require significant work) is that before the check acted as a filter on the originally matched object, whereas now there's a transform/translation.