The language of the R.gveda shows many circumstances where description is simplified if one can invoke the distinction between heavy and light syllables, where a heavy syllable contains either a long vowel or a coda consonant. In all such cases I am aware of, any sequence of two consonants parses in such a way that the first one closes the prior syllable, confirming the CS hypothesis over against the OM hypothesis, which says that at least some combinations of consonants go together to form an onset for the following syllable. On the other hand, a single intervocalic consonant is not enough to make the preceding syllable heavy, so it must regularly syllabify with the following syllable.
There are however some weaknesses in this demonstration. The first is that the data are probative only for the first consonant that appears after short vowels. If the vowel is long, the syllable is going to be heavy, regardless of how the following consonant or cluster syllabifies. Furthermore, a syllable closed by one consonant is indistinguishable in weight from one closed by more than one consonant: a cluster of two intervocalic consonants could conceivably both go with the prior syllable. This does not seem likely, however, since a single intervocalic does clearly go with the posterior vowel: one may generalize that the last segment of a cluster always goes with the posterior vowel.
The second weakness is that only the metre is an exceptionless phenomenon that is unambiguously phonological in the Vedic language. The other phenomena all are lexical phenomena which one could argue are inherited from an earlier stage of the language where perhaps other syllabification rules applied. Occasional exceptions highlight this possibility. Thus one finds the perfect véttha `know' (RV 10.15.13) despite the rule that other heavy perfect stems ending in a consonant add i before a consonantal ending. And one finds davidhâva `we shook' (RV 1.140.6) despite the rule that other intensives have long î before single consonants (dh is a symbol for a single consonant, an aspirated stop). And Sievers's Law in the R.gveda is riddled with exceptions, morphological conditioning, and plain optionality: Hock (1980) considers it to be a set of variable rules in the Labovian sense (but see Kiparsky 1972 for a bold claim that it is actually exceptionless), and certainly many words can appear either with glides or with vowels, apparently ad libitem (or metri causa). Nevertheless the other rules are close to exceptionless, and at any rate I know of no situations where an alternation sensitive to syllable weight systematically fails to treat any set of two-consonant clusters as if they fail to make position, even universally favoured onset clusters such as tr.