The simple overlapping Venn-like diagrams that people use to express the relative power of schema languages can mislead as well as inform. Instead, here’s a kind of diagram with shaded edges: light grey represent capabilities that are possible but inconvenient, and darker grey represent capabilities that are possible but only if you organize your schema in a certain way. XSD extensibility is an example of the darker grey capability: unless you organize your (base) schemas properly, you may easily find extensibility is not possible.
It is very common to compare schema languages using a Venn-like diagram with concentric circles showing, for example, that XSD grammars are more powerful than DTD grammars but that RELAX NG grammars are more powerful than XSD. Or that RELAX NG provides more powerful construction capabilities than DTDs, but that XSD provides more powerful capabilities than RELAX NG. Or whatever the author wants to show. (Then Schematron comes along, and people rarely know where to draw the line.)
A good example of this simpler kind of diagram is in Janus Dam Nielson’s very readable master’s thesis Relations between Schema Languages for XML at page 52. I recommend any student wanting to get a grip surveying the capabilities of schemas and the classes of grammars involved should start with this thesis: well done Janus (and Anders!). The thesis surveys different schema languages with a nice approach of seeing how well each supports the kind of underlying concern of each (Schematron, for example, really is based on the notion that expressing the constraints in natural language is the entry point to constraints: humans first, computers second: to what extent do other schema languages support this?) , looks at the capabilities of each language, and has a really nice section on inter-translatability.
(Note: Slightly revised)