I think it's because the thumbnails are small, so if the faces aren't exaggerated you won't perceive any particular emotion at all. It's a bit like the difference between theatre and film acting.
This is probably why. Similarly, stage makeup on actors is typically very heavy to accentuate the facial features at a distance, making them easier to read.
But that starts from the premise that you even care about the "emotion" expressed in the video. While I'm sure that there are many genres of videos where the emotions expressed by the authors are relevant to the topic at hand, there are many topics in which an expression of emotion, let alone an exaggerated one, is completely out of place.
I don't think that's true. I'm sure a lot of youtubers want people to enjoy their videos and get something from them, but they're also monetizing eyeballs. That means clicks on videos, which means grabbing attention in the video thumbnails. All other things being equal, people will click on a thumbnail with someone with an exaggerated facial expression more often than a neutral one.
As to why that works, my theory is that we're wired to respond to more extreme emotional displays because we're socially oriented great apes and that kind of information is important for navigating the world.