As with most theories that are not immediately and obviously wrong, the only way to prove it (I think) is to provide a reproducible decoding into sensible plaintext for a large enough part of the manuscript. And the only way to disprove it is to provide a similar decoding but using some competing theory.
Generally this system may work and I think it has been discussed in the past. Also, it's reminiscent of Cangjie or Wubi input methods for Chinese characters. For example:
我不會寫，但我還是可以試試看
HQI MF OMWA JHXF OAM HQI YWMRV AMYO MNR VIO YRMPI YRMPI QBU
I think this may look a bit like Voynichese under some simple substitution.
Generally this system may work and I think it has been discussed in the past. Also, it's reminiscent of Cangjie or Wubi input methods for Chinese characters. For example:
我不會寫，但我還是可以試試看
HQI MF OMWA JHXF OAM HQI YWMRV AMYO MNR VIO YRMPI YRMPI QBU
I think this may look a bit like Voynichese under some simple substitution.