"Our results have shown that with the exception of trivial binary encoding strategies, most binary encoding optimizations yield performance improvements in only limited applications or situations, and/or restrict the ability for pipelined XML processing. This supports the contention in  that there is not one binary encoding standard that could satisfy the needs of all applications. On the contrary, however, a trivial binary encoding standard would appear to at least provide performance benefits to most applications, without any significant drawbacks other than compromising the view-source principle."
Interesting paper. I think this is what those of us who oppose binary XML understand intuitively, but it's nice to see the emergence of research along those lines.