I'm not optimistic about this at all, but just for sake of methodology: I'd advise you treat Currier A and B as separate "conlangs" at the very least, you could go further and split the corpus according to RZ languages even. If you're still hopeful about matching a conlang like this, you'll also have to tackle the grammatical differences between Voynichese dialects and natural languages, taking into account the distribution of f p in the text, start and end of line strings (and LAAFU itself), horizontal and vertical reduplication, etc...
Try not to brush mismatches off, keep track of them as a sanity check to get a feeling of the quality of your results as a whole. Matching 'dio' in the right sections doesn't mean anything if it pops up in nonsensical contexts, or if the resulting grammar doesn't make sense even if the word-for-word mapping is undone. The grammar need not match exactly the patterns of Latin, but if it's a conlang you can expect some internal consistency.
Try not to brush mismatches off, keep track of them as a sanity check to get a feeling of the quality of your results as a whole. Matching 'dio' in the right sections doesn't mean anything if it pops up in nonsensical contexts, or if the resulting grammar doesn't make sense even if the word-for-word mapping is undone. The grammar need not match exactly the patterns of Latin, but if it's a conlang you can expect some internal consistency.