TAOCP is a very good reference book, and I think very few people would have ever read it cover to cover, but it is good if you want to look up an algorithm. For example, I went through the hash table chapter while doing assignments for a course while trying to optimize my implementation.
I don't think Knuth put those together expecting people to read through them like a novel.
They are, however, an excellent reference when you are faced with a practical algorithmic problem in a real world production situation in that they will often yield a direct solution to your headache on a platter and allow you to carry on with your day.
I owned a copy of this, but I only used it as a reference once or twice. Sadly it didn't survive my emigration.
I don't think I've ever met anybody who has read the whole thing, and if somebody told me they had I'd assume they were lying.