This is a rather different case of ambisyllabicity than the familiar Germanic ones. In those languages ambisyllabicity probably arose because of a drive to align stress and moric prominence: i.e., stressed syllables had to be heavy. Therefore a single consonant after a short stressed vowel had to be geminate. The fact that gemination was largely redundant made it particularly susceptible to phonetic loss, leaving ambisyllabicity as its structural remnant (Kurylowicz 1948:208). In Sanskrit, on the other hand, stress is not involved,Hulst 1985:65). A similar argument would apply to a hypothesis that Sanskrit attempts to balance a maximum coda requirement against the maximum onset requirement: singleton consonants should not be immune.
The most obvious explanation is that the main motivation for the ambisyllabicity is the consonant cluster itself. A cluster of two or more consonants wants to be bisyllabic at all levels of the grammar, even in contexts where there is a separate constraint for maximum onsets. In these cases both constraints can be solved by means of ambisyllabicity, since all relevant consonants can be found in codas, especially when linked to a following onset.(12)
I would further speculate that the specific property of clusters that makes them want to be bisyllabic is weight. Here I conceive of weight not as a property peculiar to syllables but as a feature inherent in segments that is inherited by the constructs they are found in. In general segments have one mora apiece, although geminates have two. In a construct, moras are summed up to the limit of two (reflecting a general tendency in phonology to make binary distinctions), so segments can make a rime light or heavy. A syllable is headed by the rime, so it inherits only the weight of the rime, not that of the onset. But the weight of the consonants in the onset could still be significant for certain other purposes.
That heavy segments (i.e., geminates) want to be bisyllabic is a universal (Clements 1990:321-322). Sanskrit generalizes this constraint to heavy interludes in general. Thus Sanskrit distinguishes between light and heavy interludes, where a light one is a single, simple consonant, and a heavy one is a geminate consonant or any cluster: light VCV, heavy VC:V or VCC+V. The language does not permit any distinctive length among elements in branching interludes (consonant clusters).(13) This is exactly analogous to a distribution often found in rimes in languages like Pali (Hermann 1923:260), where there is a distinction between short and long vowels, but not in branching rimes (closed syllables).
Why should Sanskrit have a more general heavy-interlude constraint? Possibly there is a connection with its verse form, which is sensitive to weight. It may be the case that languages select as their fundamental verse structure their fundamental timing unit. In Germanic languages, this is the stress foot; in Spanish and French, this is the syllable; in Sanskrit and Ancient Greek, this is weight. Conceivably languages timed by weight are particularly sensitive to the weight of interludes. A general requirement that heavy interludes be ambisyllabic would tend to make the weight of clusters metrically relevant, since that would make preceding light syllables heavy.
It will be noted that all documentable cases of ambisyllabicity in the R.gveda involve onset clusters, so one may object that I should be talking about heavy onsets instead of heavy interludes in general. The distinction is important. Onsets may not generally be considered to have weight (inasmuch as they do not influence the weight of their syllables), but at least there is a strong consensus that they exist. Interludes are not generally given independent status in current phonological theory. Nevertheless structuralists such as Kurylowicz (1948) found it useful to speak of the structure of intervocalic consonant complexes. And evidence to be presented in the next section indicates that further developments in Indic cannot be based on onset weight alone. Interludes are clearly not part of the prosodic hierarchy, since they do not align consistently with the codas and onsets of syllabls. Rather, they should be thought of as a sort of presyllabic organization of segments on a separate tier.