Voynichese is a numeric cipher?

[ololololo]

I don't think it has even been a problem to find a way to convert plaintext snippets into Voynichese snippets without any difficulties. And if nulls are involved it has never been a problem to explain the how of 'chol chol chor' (not the why though - why make the nulls obvious this way). The problem as I understand it is to match the large scale statistics of the manuscript.

Also, if your method involves sorting letters in each word, I think it's not encryption, because the underlying text is mostly destroyed. There is no way to tell 'read' from 'dear' or 'news' from 'Swen', it's just a way of generating garbage from a plaintext.

You're right! It really is. Therefore, I assume that the author provided clues for decoding, expressed in the form of repetitions and similar words.
The problem with READ and DEAR is actually solvable. Suppose A = I, E = II, D = III, R = IV. R-E-A-D = IV-II-I-III, from more to less: IV-III-II-I (RDEA); D-E-A-R = III-II-I-IV, from more to less: IV-III-II-I (RDEA). It's really unclear, and the correct word can only be guessed from the context. But I can do this: III-III IV-II-I (DD REA), or III-III II-I-IV (DD EAR), and it won't be so difficult to distinguish DEAR from READ. We can easily mark the first letter or simply leave it in place.
I assume that the author had basic intelligence and was also aware of this problem, but he could have solved it differently...