The Voynich Manuscript is a cipher because...?

[Emma May Smith]

Assuming that the author had a plain text, either in his head or in writing before him, and he somehow converted this to the Voynich MS text...
Then the low entropy either was already there in the plain text, or it is a result of his operation.

Looking at this operation, the random insertion of null characters is more likely to increase entropy, unless it wasn't really random but followed some pattern, in which case it isn't fundamentally different from a verbose cipher.

Compression algorithms (like the well-known 'zip') specifically increase the entropy, and the uncompression decreases it.
To decode the Voynich MS text, one is then looking for something similar to compression.
Removal of nulls would be a form of compression, and combining characters another.

Things I have seen in some translation attempts, e.g. suggestions that the gallows characters represent the Nahuatl character pair 'tl', are going into the wrong direction.....
Another suggestion: that the text is vowelless, and vowels need to be inserted, is equally going in the wrong direction. Such a further reduction of entropy would need to be balanced by 'something else' that stronly increases it.

Surely the answer, or at least part of it, is that we're deconstructing characters, or character combinations, too much?

I know that the issue over [ckh, cth, cfh, cph] has been mentioned before, but also [ch, sh], [e, ee, eee], and [in, iin, iiin]. What's the highest entropy we can reach just by making different (but reasonable) character identifications?

Also, what improvements would it make if we assumed, for the sake of argument, that some characters had more than one permissible reading? The readings could be linked (such as voiced and voiceless) and distinguished in speech though not in writing.