How many pages?

[MarcoP]

I am not sure this has been mentioned yet. Though I can't find the original post at the moment, I think it was Rene who pointed out that solvers should focus on the encoding process, rather than decoding. They should define a repeatable encoding process and encode a couple of short historical plain-texts in the source language; it is then possible to check that the results feature all the properties of Voynichese,  e.g.:

• low character entropy
  
• zipfian distribution of words
  
• frequent repetition of identical words
  
• even more frequent consecutive similar words
  
• binomial distribution of dictionary word lengths
  
• constrained paragraph distributions for p f
  
• paragraph initial words with a preference for starting with t k p f
  
• constrained line distributions for m g
  
• words that mostly adhere to Stolfi's grammar
  
• a tendency for final y to be followed by the q prefix
  
• 'dialects' with largely different bigram distributions
  
• ...add you own favourite Voynicese properties
  

I think that doing this on two 500 word texts should be enough for an initial evaluation. This could possibly be equivalent to Nablator's point about the predictive power of a theory.
This method totally removes the problem of the meaningfulness of the source text. One encodes historical texts which are known to be meaningful and grammatical.