threads are placed on a queue, like in any other system, and they are run by the idle processor based on their state and prioraty.

If it actualy happens that the processor will trigs a fault (as you say, I have not read CnQ whitepapers) everytime it needs more power. That fault will be managed in a ring0 (kernel mode) trap handler. Which by nature will put the thread it was running back in the queue, like any other system supporting SMP with premptive scheduler. And if the other processor is idle, or running a thread with less prioraty, or even running something else, but for enough cycles that the system thinks it is better to switch to another task. It will take that paused task and run it.

It is not actually vistas fault. But AMD faults not considering how proc schedullers works. If they managed to make it and opt-in feature, that OSes will turn on if they manage processor prioraties based on their clocks or handle clock scalling in a better way (in a non interrup call, or with a low prioraty process on the second proc). it would be a welcomed feature for sure