The W3C XML Binary Characterization Working Group has published the second working draft of XML Binary Characterization Use Cases. This divides roughly 50-50 into things that should be done in plain vanilla XML (Web Services for Small Devices, Web Services as an Alternative to CORBA, Electronic Documents, FIXML) and things that should not be done in anything remotely like XML (Floating Point Arrays in the Energy Industry, PC-free Photo Printing). In brief, they're trying to turn a station wagon into a Ferrari, and instead they're going to end up with an Edsel. Despite the hype XML is not, cannot, and will not be all things to all people. At best this effort will fail. At worst, it will fail and take down XML with it.
A few of the use cases (Embedding External Data in XML Documents, PC-free Photo Album Generation) demonstrate legitmate needs to bundle binary data with XML. However, they're doing it inside out. The XML and the binary data should be combined in a non-XML envelope like XOM proposes, rather than forcing the binary's square pegs into XML's round holes.