I've found the author of Voynich manuscript. He was Giovanni Aurispa.

[tavie]

My interest is why so many independent approaches all work (read: fail) in the same way.

Yours and Koen's similar question is really quite thought-provoking.  It would probably make an interesting thesis to examine all the factors behind this, and it would also be interesting to see what the gender balance of solvers is.  Ultimately, though, I think the reason that all wrong solutions are wrong in the same way (to misquote Tolstoy) is because all "solvers" are caught by the exact same trap at the very start.

At the very basic level, I think the two biggest factors are:

• a) not knowing how easy it is to "see" words of any language in the Voynich.  If you don't have experience with statistics or comparative linguistic, it's hard to accept that even one single word is just a coincidence.  Intuitively, it feels that it can't be a coincidence. 
  

• b) wanting to be the person who deciphers the undecipherable manuscript.  This combines badly with a).  Thirty minutes' worth of reading failed theories on this forum is enough to fix the lack of knowledge above.  All those people with different solutions claiming their findings "cannot be a coincidence" is an unbeatable demonstration of how common coincidences are.  But solvers are so keen to crack open the mystery that they dive straight into developing a system without doing their reading.  And once they've come up with their solution, it's too late because they are now personally attached to it.  Confirmation bias waves away and minimizes the problems. 
  

Imagine you didn't know how common these coincidences are, and you've just heard about the Voynich, and you think "Hmm, I bet no one's ever looked into it being Gaelic", and so you crack it open, spend a few hours toying around possible letter matches for the glyphs...and then you look up an online Gaelic dictionary and one of the first meanings you see is "plant"...it must feel like you've struck gold or won the lottery.  It can't be a coincidence. 

And so you try out those letter matches on other words, and with a bit of trial and error you find words that also could belong in the Voynich, e.g "heal", "find", "ladle", and you build your 1:1 alphabet. It wasn't as easy as your first match.  Sometimes you have to scroll to maybe the sixth alternative definition of a word, and you may have to alter the word slightly.  But it was there in the dictionary, and you already found "plant", so this is solid proof that you have found the right system.  

Oddly, after finalizing your system, it gets harder rather than easier to read more words.  You find yourself having to alter the words a lot, and people point out this gives you too much freedom, that it doesn't work with Voynichese properties you've never heard of, that it doen't make sense in Gaelic. But you know your earlier results were not coincidences, and so whatever these problems are, if they exist at all, they must be only minor issues that can be explained somehow.

Basically, I think that initial moment of discovery is doing a lot of heavy lifting, and people won't change their minds after it's happened, no matter what.  If solvers could evade this giant trap at the start, we might see wrong solutions of a different type, but they don't.  The same phenomenon was seen for Linear B, and we just see more of it because the VM is so accessible.  Short of adding in a "Coincidences happen!" warning before anyone can access a folio, wrong solutions will keep being wrong in exactly the same wrong way.  Which is quite depressing, when you think about all that lost potential contribution.