Those are exactly the questions I ask myself every time I hear about a new, super-fast XML parser. Is it a real XML parser[1] or can it only handle a "simple" subset of XML? I am surprised that more people don't seem to share the same skepticism as you and me. Probably because most people don't realize that implementing a conforming XML 1.0 parser is not a trivial task.
And you are right about the XML data binding tools. In most cases parsing XML itself is not what takes the bulk of the time. It is validation (against DTD, XML Schema, or ad-hoc), data conversion (e.g., from "123" string to 123 integer) and perhaps construction of some in-memory representation (with memory allocations that it involves) that take the bulk of the time. There are existing tools[2] that can performs the above tasks an order of magnitude faster than a general-purpose XML parsers by generating the tailor-made validation and data extraction code as well as data structures from XML Schema.
[1] http://www.codesynthesis.com/~boris/blog/2008/05/19/real-xml...