Still, for me, if the text produced with the Cardan grille wouldn't have such statistics, that would be a proof for me that the "language" is real. That's what I hoped to find. If the Cardan grille generated text would pass the statistics too, I'd still more believe the text to be a kind of hoax.
The game isn't "lets find which statistics Voynich passes" but "let's disprove the hypothesis it is a hoax" by finding a test which all claimed hoaxes schemes obviously fail.
3. Why do you give hoax higher prior probability?
EDIT: I was wrong about the possible distributions so there's no point 2.
3. Because nothing else "fits" with anything else we know of. When we find first text in an unknown language, typically it's some religious thing. The illustrations are at best "dream" constructs when not intentional attempt to make an "original" hoax. Otherwise, it's obvious that it's not something out of some non-western culture.
Even if the overall goal was some kind of hoax, the text itself almost certainly carries semantic meaning, and that in itself is fascinating due to its apparent indecipherability.