Ok, let me expand.
I understand that part of xorlo is that predicates be defined on plural variables. This allows, for example, {bevri} to be defined in a way such that {lo nanmu cu bevri lo pipno} can mean that the men carried the piano either individually or as a group. This does not mean, though, that the predicate has a double meaning, it just happens that it is _cumulative_. That is, if one man carried the piano, and the other carried as well, then we can say that the two of them carried the piano as well, so that the distributive meaning is particular case of the collective one. If the predicate were distributive, the converse would also be true, and no distinction would exist.
Defining the predicates as collective is justified because distribution can be easily conveyed via external quantification and we don't have to introduce any artificial mass-entities otherwise.
Well, it turns out that cumulativity is very very common among properties. It wouldn't be hard to enumerate the non-cumulative properties. {cmamau} is a good archetype. Other examples are {cmalu}, {ci mei}, {me'i ci mei}. There is not much space for creativity beyond that.
It is to be expected that it brings some contention.
To me, nothing else having been specified, the most natural thing is to give {cmalu} and friends a collective meaning. Natural enough that the sentence in question caught my attention during a fairly fluent reading. We can still distribute via external quantification and there is {lo'e} for myopic singularization.
I agree with xorxes' myopic singularization analysis of "The men were smaller than the adults Dorothy was used to". I just don't see a justification for that in the Lojban citation.
Allowing all predicates to have either collective or distributive, or generic interpretations, without clear rules about when each is to be meant, is to me just bad ambiguity, no better than ambiguous scope, or allowing {lo broda} to possibly mean {lo du'u ma kau broda}. Once we can formally fully explain natural languages, we can have a LoCCan just like them (though there would hardly be any point in it).
I can envision some ways to systematically justify the distributive interpretation. {cmalu}, {barda} and friends could be defined to distribute over its connected components, for example. That would capture xorxes intuition that "comparing people's sizes is more common than comparing groups of people's sizes" while allowing the collective meaning in convenient contexts. I would still want something like that to be officially and explicitly declared somewhere, though.
mu'o
mi'e .asiz.