It's an interesting puzzle, sort of like a crossword puzzle. I'm certainly open to the idea that it's a hoax until someone actually figures it out. And if so, it will probably be a lot less interesting than it is now, as an unknown.
The most compelling argument about it currently is that it has attracted so much attention yet remains unsolved. That's the strongest evidence we have. (I'm not quite sure what it tells us-- maybe it's an unknown language, or an inconsistent code, or a hoax, or various other possibilities.)
You might be interested in this other recent discussion on the topic here:
http://linguistforum.com/outside-of-the-box/medieval-manuscript-reveals-proto-romance-language/I wrote quite a bit in response there, but it would equally apply to any other theory. The bottom line is that I'm skeptical about any proposal until it reaches the critical point of actually transparently speaking for itself and with consensus from other researchers. Who knows, that one might be right. But I won't believe it until it gains some ground in the community especially from other people who have studied the manuscript and realize their own ideas are wrong because another idea clearly gets it right and can fully and transparently translate the text.
As for specifics, I don't know enough about it to comment on the rest (it's mostly just speculation at this point anyway), but I fully agree with your point about Zipf's "law". It also applies to basically any other sort of list-like data, such as names in a phone book. Although the linear distribution after logarithmic modification is perhaps surprising (though we don't always find exactly that anyway), the fact that some items are more frequent than others should not surprise us at all.
