Is there some way to assert "Given the same audience of X people, adding such thumbnails results in more clicks"?
And it isn't entirely true that including a face mugging for the camera is always the most visually engaging (although representations of people are very attention grabbing). A person in the image is a character for a story, the whole "worth a thousand words" is true in that a whole narrative can be compressed into a single image and viewed with a glance. A person or other sentient being is a character for that story (I'm sure cat thumbnails do well too).
Just try to think of an interesting narrative involving exclusively inanimate objects. Kinda hard; the whole "animate" thing seems to be necessary to give a before, middle, and after to events (ok, maybe collisions, explosives, and rockets might work, they'd probably do well as thumbnails too). A thumb with Linus looking surprised or disappointed or puzzled gives a short and incomplete narrative about the object he is looking at and how it made him feel those emotions. Part of this is that watching the video will give you a more complete narrative.
I would think a human face would be the most effective thumbnail, and that there are psychological reasons for this. When I worked in print magazines, there were metrics thrown around a lot for how well cover subjects did in terms of newsstand copies sold: Animals > inanimate objects; Humans > animals; color > b&w; photos > illustrations; eye contact with the camera > looking away; females > males.
Apps are also getting into the fun, and tons are now including screaming faces[1] as their icon, no matter how related to the gameplay. Can't wait for a productivity app trying to pull off this icon...