The ACPI _CSD maybe unworthy of our trust
While debugging #1627 I happened upon the fact that (it appears) we're the only OS to actually trust the ACPI _CSD (actually, the only to use it for anything). Due to this, it seems to be not at all uncommon for it to be utterly incorrect, to the detriment of Power Aware Dispatching, etc.
It is, in almost every case, more useful to just assume that CStates are per-core. An exception to this may be on the new AMD Bulldozer CPU's, which may have a concept (not unlike the procnodes of Magny-Cours?) whereby two cores are part of a larger container, and power-manage together (that is, the CState domains are one level higher, at this new level).