What if Voynich plaintext is very heavily shortened?

[nablator]

Voynichese as a whole is not a consistent system though, so wouldn't this argument work against most cipher-based solutions? We're always dealing with one subsystem shifting into another, or certain preferences dominating different sections.

Most cipher-based solutions, yes. Some proposed solutions are:
1. Some parts are nulls. Pig Latin and other verbose expansions are possible. In steganography most of the text is filler.
2. Variable settings cause differences in statistics.
3. There are multiple plaintexts for the same chunk of ciphertext (stateful system).
4. Meaning, if there is any, is spread over larger chunks of ciphertext than glyphs and glyph bigrams (trigrams, or words: then Voynichese it is not a cipher but a code).

My preference is 2+3+4.