Why not positional variation?

[Barbrey]

I also would love to see some research on that, although to be fair, even with such kind of substitution (e.g. accounting for spaces as possible characters in n-grams), I'm betting the text won't exactly match a natural, romance/germanic language plaintext due to

• LAAFU/PAAFU
  
• reduplication of (long) words
  
• reversibility of vords (there is some evidence that suggests Voynich word order is far more flexible than any natural languages. Talking I'm about this kind flexibility of.)
  

But who knows really. It could very well be a mixture of many ciphers, including substitution (assuming of course, the text is encoded plaintext and not gibberish)

True. I think your examples only support my speculation though. Word length, duplicates, reversibility - none of that would matter. In my example, only the first and last letters of each word would matter, meeting in the space. The space doesn’t have to be decoded for statistical purposes, it can just be a binary glyph.