The flattening of users could also be due to a general internet-wide reduction in long-form (or even medium-form) non-fiction reading. How are page views for The New York Times?

Seems like it should be simple to A/B test, though. Obviously Google could do it themselves by randomly taking away the widget, but would could also see whether referrals from non-Google search engines (though they are themselves a tiny percentage) continue to increase while Google remains flat.