The premise that grammatical theory should be responsible for predicting diminishing probabilities of occurrence is a problem. All you can do is generate / not generate a class of structures; so you'd have to identify something about the class that distinguishes it from other structures that are generated. What you can do with just C and V is very limited.
You should look at the original source of the technology, Clements & Keyser's CV Phonology: a Generative Theory of the Syllable. Their theory starts with the concept of a "core" CV syllable, and transformations on them. One is allowing onsetless syllables -- that means that no language has obligatorily onsetless syllables. Another is allowing syllable codas. This means that no language could have only onsetless syllables, or only closed syllables. Some languages allow any number of vowels in a syllable, and they allow "*" meaning "one or more" to follow V and C, and in fact allow a specific integer value for any instance of *, thus a grammar could allow a maximum syllable of the form "32 consonants plus 6 vowels plus 12 consonants". However, as far as I know no case has ever been made for setting a specific numeric limit on consonants, since observed maximum sequences always follow from a separate principle.
Other aspects of the theory address the featural content of syllable-initial and -final sequences, such as the fact that English onsets can be [st] but not [ft]. This is the mechanism that drives most of the power to rule out odd i.e. "unlikely" clusters like [mk]. In a later paper ("The role of the sonority cycle in core syllabification") Clements develops a general theory for encoding sonority generalizations, but there will still be language-particular restrictions on certain sequences. The C&K theory has two sequencing mechanisms, positive conditions such as that onset clusters of obstruent plus glide / liquid (tr) are allowed in English, and negative conditions that non-strident coronals cannot be followed by a lateral (tl). Using these kinds of conditions, one can actually allow in unbounded set of onset and coda consonants in English, and derive the observed limit (CCC in onsets for instance) from the impossibility of parsing CCCC while obeying these sequencing conditions.
Returning to "*" in core syllables, it does appear that vowel sequences are different, in that there are languages that allow just two Vs in a syllable. Hawaiian would be an example. Within CV theory, one would say that Hawaiian has a (C)V2 syllable template, where V2 means "up to two vowels", at least if it is true that Hawaiian doesn't allow more than two vowels in a syllable. (I've never seen a phonological argument to that effect, it's just stipulated). In that case, hooiaioia 'certified' (documented as a Hawaiian word, not just an internet meme) would have to be more than one syllable -- 4 syllables, perhaps more. Let us suppose that there is no word with the form CVVVVVVVVV, which would have to be at least 5 syllables. But there are 9-syllable words like kawaiolaonapukanileo "choral group", so there cannot be a word-size limit that words have to be under 5 syllables. The notion of morpheme- or word-structure conditions fell into disrepute and there isn't a formal theory of them, so it's hard to tell how one might state such a limit -- at any rate, it would not be a limit pertaining to syllables, it would be about phoneme sequences.
There is no word "plagrafufu" in English. There is no reason to elevate that gap to the level of a matter of formal principle, i.e. rule it out. Likewise, there is no reason to specifically rule out apparently non-existent hooiaioiai in Hawaiian. Especially in languages with long words, the set of "actual" words in a language is often a small subset of those generated by motivated syllable structure principles.