Voynichese is a numeric cipher?

[ololololo]

I don't think those (rare) repeated labels need an explanation.

Well, in the case of a plant and fire, this is quite possible, but this word also appears in other contexts. There are other words similar to this one, for example, otol. 

One could, of course, assume that such frequent words are common and typical for the context, but that wouldn’t change anything.

Synonyms: two different words with the same meaning (like buy and purchase).

.
We don’t know whether Voynichese can be pronounced, so we only have a difference in meaning.
In this case, I used synonyms in terms of “similar words.” I admit, it was my mistake 

To decode the text one would need a copy of the cards.

First, it needs to be somehow made...

This would recover the original plaintext exactly, so homonyms etc. would cause no more problem than reading the original would.

The author has an advantage over the addressee — he is the creator of this cipher. Perhaps he has simply memorized the frequent word forms, and overall, the ciphertext is perceived by him more easily and quickly. The addressee, on the other hand, will have to dig through the cards and decipher the text word by word. This is where the point of inefficiency arises.
However, I’ll note here again that this is not advisable for us. Perhaps the recipient wouldn’t be too lazy to decipher it (in general, they could spend some of their time to create something akin to a copy), and the author wouldn’t be too lazy to do it.

with an empty box, the first few cards could be...

In your example, the words evolve from simple to complex and combined. It seems that @npcompl33t found some kind of word evolution, but I don’t really understand whether this can be interpreted in the context of a book cipher...