We're talking about Knuth's opinion.
The whole point behind Literate Programming is that you can't read non-literate code as literature, and Knuth believes being able to read code as literature makes it easier to understand. If you could read any sort of code as literature then why would we need literate programming? So given that virtual all code in existence is non-literate it's pretty obvious that trying to promote code reading as literature reading is not going to work that well. The point of this article is that the author tried it both ways (code reading groups reading code as literature and treating it as an investigation) and found the investigative approach worked better. He's not expressing an opinion or proposing a hypothesis, he's simply reporting what worked and what did not.