The first and second and third documents are equivalent; only in terms of the hierachal relationship between nodes. They are not by any means equal, and depending on the actual XML language they belong to, might not be considered equivalent semantically; nor are they necessarily equivalent in terms of hierachy if nodes are considered instances of something as opposed to types of something. That is, you might be saying, in the first case, that the X's in A are constituent parts of some single 'definition' of A, whereas in the second case, you might be saying that each X within an A or B is a unique X 'instance' within some unique A or B 'instance'. It depends on the XML language or representation, but in most XML languages I've used, the treatment of nodes is as instances, and the order of nodes that are in a 'sibling level' sequence is considered to some degree; so, the first, second and third examples would certainly not be considered equivelant on any meaningful level in a language like say, XHTML or XSLT.

In short, if you're authoring the XML dialect, it's up to you; it's merely a design decision. If you're working within the bounds of an existing dialect, you have to consult the specification of the dialect, because there's no way you can assume any level of equivalence.