Far from conclusive evidence here. And my intuition says something different to me. A controlled experiment would involve showing the mobile UI on the desktop and vice versa. You might also want to control for things like page load time (native mobile apps are often snappier)