To be fair, this is pretty much the norm in all of academia. Universities traditionally existed to learn things, not to make useful things out of them. Turning a research project into useful code generally gets left to external entities or spinoff companies (Xen's perhaps the most notable example of this in the Linux world).
This doesn't excuse the fact that a great deal of CS research seems to entirely ignore real-world constraints - coming from a (non-CS) academic background, I try to keep vaguely abreast of relevant published work. And pretty much everything written on power management has been based on assumptions that are either untrue or entirely irrelevant in the modern world, or alternatively has had an implementation proof that turned out to work entirely by accident.