I started believing that Googlebot would render entire pages, when Google Pagespeed was introduced. When they calculate page loading times, they can also produce a waterfall view of the rendering progress.
Before I suspected some rendering/css/DOM parsing was used to detect hidden text and links, but not on such a massive scale. Like now you can index a new page and it almost looks like a preview image gets generated "on-the-fly" when there is no cache available.
I wonder how much of this render data is used for Google Panda quality control updates. Position of advertisements, call-to-actions, author information, address information and disclaimer/privacy policies could now all play a role in making your site seem more credible in the eyes of Google and your users.