One simple example is with copy and paste. NLS had a highly complex and error prone copy-and-paste system. A lot of the commands in NLS were modal (vs modeless), and thus error prone. See this article for more info: http://worrydream.com/refs/Tesler%20-%20A%20Personal%20Histo...
Another issue was basic usability of NLS. Engelbart's vision was on continuous bootstrapping, continually aiming to create more powerful and expressive systems. This vision was pretty good for technical people, but also comes at the cost of basic ease of use and entry for newcomers. For instance, think about how many problems non-technical people still have with today's PCs and smartphones, many of which strive for "walk up and use" kind of usage.
Having been in research for over two decades now, my position is that we researchers will often identify the right problem and get the right general direction, but not always the right form. For instance, in the late 90s at Berkeley, folks correctly identified cluster computing as the future, and that we should aim to create abstractions to host Internet services to make them more scalable and robust. The specifics went in a very different direction (e.g. cloud computing, Kubernetes, etc), but the general thrust was right. As another example, my colleagues and I correctly identified rapid prototyping and testing tools for user interfaces as a major need, and while we inspired a lot of ideas with our work (pen-based sketching), industry went in a very different direction (see Figma and Axure and InVision).
I think the same is true with Engelbart's work. He was definitely a pioneer and helped open up people's imaginations as to what was possible with computing, but was off on a lot of the details and specifics that really matter for adoption. And that's ok because to a large extent, that's one of the many reasons we as a community and society do research.