There's a debate raging about whether binary XML is a good idea (summary: binary XML would undoubtedly have performance benefits, but would it undermine crucial benefits of XML?). Ingo Rammer is the proponent of the idea and while I concede virtually every point he makes, on balance it seems to me a bad idea. While XML is amusingly wasteful of network capacity ("\<boolean>true\</boolean>"), it seems to me that the text-based nature of XML is a crucial benefit. You can read, edit, post, and consume XML with no (or extremely minimal) specialized tools. In my experience, this is a key benefit that allows different development teams to move at different rates: being able to capture and manipulate the exact request or response with an absolute minimum of system intervention is very valuable.
As soon as you introduce binary into the equation, you make it harder to say "And what if we change the dates to December 31 and January 1?" or "Is the 'count' attribute accurate?" Additionally, you make it harder to work in a disconnected fashion: Web designers can work with an XML file to design XSLT transforms; server-side developers can post queries into a textbox to mimic the behavior of a not-yet-developed client. These are very important benefits.
To be fair, the binary XML argument is "text plus binary" (essentially, add a binary parser to the equation while maintaining compatibility with all higher-level standards -- if the mime type is text/xml use the text parser, if the mime type is binary/xml use the binary parser). But then one has to maintain confidence in the processes of the binary parser, the binary producer, and how the choice is made between text and binary representations. That seems to me a big ol' can of worms.