And Google can't classify Instapainting.com as a "photo to painting" site. It just knows it's a site that contains some content. Since Yahoo.com would get better rankings even with disparate topics under the domain.
I was at a talk yesterday where the speaker from Google was saying that while pagerank is the number one input they use for search results, AI style machine learning algorithms are now number 3 so don't be so sure they don't pick up on that.
I am fairly certain that the world's best search engine is not falling for that. How can we decide who is right?
But in this case, I don't think Google was necessarily "fooled". The developer happened to be multitalented enough to create a game that was popular. Assuming that the traffic it receives was legitimate (which I imagine Google can know based on Google Analytics), how is that not a signal that the domain belongs to someone of more notability than just some random domain? The problem would be if Google used that signal as the sole determiner of whether Instapainting.com was a place to find good paintings. But there's no indication that that is the case, just that clout gained from the popular game may have helped Instapainting overall be seen as a legit domain, generally.
Ideally, Google wouldn't credit one part of the site that is lower quality than the other part.
Site authority presumably wouldn't be a thing if it didn't produce good results. Sites that allow crappy content will tend to have more crappy content and so a page on such a site is less likely to be what someone is searching for; as long as that stands then site authority will still be useful and shouldn't - IMO - be discounted completely.