Commas are weird punctuations; there are as many ideas about what they are for and how to use them as there are style books (and probably more). At heart, though, they seem to indicate a place where a speaker might take a very brief pause because it was at the end or beginning of some sort of separable syntactic unit that might not otherwise be clear (there! that should be vague enough to cover about half the cases.) In the end, though, the use of commas (and colons and semi-colons) are governed by conventions that may have little to do with these factors (though uses that clearly go against these are rare).
In a conlang, starting from scratch, we can adopt our own conventions. In tp, the convention so far seems to be not to use commas at all (with rare exceptions). In pu, however, Sonja regularly uses a comma before the 'la' of conditional expressions. This seems an odd choice -- except as a calque of English usage -- since conditional phrases are already clearly marked off and, if more were needed, the natural speech pause is after the 'la', not before it. Such anomalies are common in developed languages, but, when stating fresh, perhaps we can do better.
It is relatively easy to come up with cases where commas would help the reader (and pauses the listener) to understand what is intended, to disambiguate cases where a number of parses converge. Four cases, in order of frequency (I think), come to mind:
1. The beginning of a prepositional phrase after a verb phrase. The ambiguity happens here whenever this situation occurs (so, in most sentences) and is kept from being a problem by context and semantic incongruity. But there are enough cases where context is not enough to resolve the issue immediately (and sometimes even after considerable puzzling over it), the classic being 'ona li pan e sitelen tawa mi, "He gave me a picture" v. "He showed my movie". A comma between 'sitelen' and 'tawa' would break the modification chain and fix the first interpretation.
2. More than two words after 'pi'. x pi y z w can be x pi ((y z) w) or (x pi (y z)) w, 'soweli pi kule pona mute' is a critter of many pretty colors or many critters of pretty colors. In this case, of course, we can avoid the problem by shifting the 'mute' to immediately after the 'soweli' for the second reading. But such shifts are not always possible (or, at least, thought of in time or elegant or ...) and so, a comma after 'pona' to indicate the end of the 'pi' structure would help. To be sure, since 'pi's can be iterated, complete disambiguation may not be possible; but the problem can be reduced considerably.
3. Prepositions as verbs with modifiers and complements: 'mi tawa noka jan' "I walked to the man" v. "I went to the man's legs" (v. "I walked humanly", but one problem at a time). Heree, the solution (taking the thrid case into consideration as well) seems to be to mark the beginning of the complement, just as the DO would be marked for transitive verbs (though only with a comma, not an 'e', of course). The alternative, if you notice a problem, would be to repeat the preposition. But problems often don't get noticed by the creator, only be the recipient, so better to strike ahead of need,
4. Modals used transitively with transitive complements. There are actually several places where problems can arise here. The basic situation with, say "I permit somebody to hunt birds", are the modal 'mi ken e ni' and 'jan li alasa e waso' which can be collapsed to 'mi ken e jan alasa e waso' or 'mi ken alasa e waso e jan'. both of which have other, more obvious, readings. Separating either the permission (ken alala e waso) from the recipient (e jan) or the granting (ke e jan) from the granted (alasa e waso) from the granting (ken e jan) would help, As would, of course, using the full form with two sentences.
There are probably other systemic cases where commas might help, but first we need a cultural decision to actually use this device and then decide on the appropriate ases.