One of the things Knuth is famous for is writing a lengthy program for word counts that was equivalent to no more than 6 lines of shell script.

I highly doubt that he's wrong regarding his own programs, but it's most certainly wrong for the majority of programmers.

PS if we're going to appeal to authority, Church and Lambda Calculus were there first.

And out of the ~250 programs that Knuth wrote last year, 2-3 would have solved a similar problem that your typical professional programmer would solve.

Functional programming strikes me as most useful for software engineering. I'm not sure how that applies to Knuth, but my understanding is that most of the programs he writes are more oriented towards computer science than software engineering.