There do remain a number of cases to which this general notion does not seem to apply. Your case of the real line is one, letters seem to be another. Here the approach seems to be to start at the top, work down and then back up, I think, but I don't know just how that goes. Even with cases that fit this pattern pretty well, water, for excample, there are some problems, as you note. Water molecules don't display the characteristic behavior of water (as do not also ice and steam), since they don't flow, etc. But then, gold atoms don't shine and are not malleable, so this seems a minor problem. And for generic cases, they probably are not significant, since the far more numerous and visible sub bunches will take over the "statistics".
I find you idea of an aspect difference among the various uses of kinds (max. bunches) interesting, though I still tend to think of
them in terms of different connections to predicates, a relic of the early days of plural reference. And, indeed, even with aspects, some of this will still come into play with regular bunches, that is to say, bunches which do not claim to take in all the possible "atoms" (I am used to your usage here).
I kinda like this result, since it leaves basic things basic but covers kinds and masses economically from them. Until some real snag comes along.