vague thoughts for an approach

[ReneZ]

In your second point, you are approaching a key question which does not yet have an answer, but I think that it has a likely answer.

This is:
Is it possible to make a translation table from words in Voynichese to words in some plain text language, and do a word-by-word translation that makes sense?
(This includes the case where there is no similarity between the Voynichese and the PT word).
In other words: could a dictionary be made?

The consequences in case the answer is NO are enormous. It means that the vast majority of approaches that people have been taking are ruled out. No substitution cipher for example.

This is hard to prove, and the single word entropy value for the text is normal enough to keep the possibility open.
The number of Hapax (words that appear only once) is rather unusual, but the main problem is the lack of repeating sequences.

To answer this question, all the analyses about word structure (which I think are valuable) play no role.

Identical word repeats (which are relatively high) do play a role and it is another argument against it.

Similar-word repeats (which are also high and certainly unusual) would not play a role.

Torsten Timm's text generation method, which is proposed as an argument against a meaningful text, could be used in two manners.
1) It is used word after word when creating a text. In this case there is no meaning and no dictionary.
2) It is used to set up the dictionary and then the dictionary is used to translate a text. In that case there is a dictionary.

As you see, this question has a lot of implications, and there is plenty of room for more thinking.