Work-sample questions is not free from its own evils. They are expensive to device. At big companies, significant portion of candidates dump out their interview questions on Internet so that's a big challenge.

Second, work-sample type of questions are also expensive to answer for candidates so you can probably just ask one question as part of whole interview and that's about it. So any conclusion is drawn from sample of 1 as opposed to sample of 5.

Third, work-sample type of questions requires knowledge of engineering skills that candidate may or may not have developed well at point in time. At many companies, emphasis is how can we solve problem X from algorithms perspective and engineering is just something you learn on the way if you haven't already. Key is ability to figure out computer science part rather than language and tooling part.

Overall, I think many companies blindly try to follow the model of Google, Facebook etc and fall on their face. If you expect your developer to write CRUD apps or do mostly plumbing work, there is no point in asking questions like find common ancestor in binary tree. At other companies, developers are expected to solve computational problems and engineering/plumbing/CRUD is small part of the solution pie. There it makes a LOT of sense to ask candidate to solve problems that requires deep computer science. Also asking as many of as possible so that your sample space is large and outcome has more confidence.