That is to be expected. It's the same with Intel. When you enable CnQ or speedstep, it takes *time* to change states. It isn't a lot of time, but that delay will reduce performance. It is for this reason I don't use those features. I set my clocks and voltages manually in the BIOS before booting based on what I am going to use the system for. This way I get the best of both worlds.
It is simple. You lose 10% performance while you save 50%+ in power consumption when idle. Idea behind power saving is to lose a little bit in performance while saving a lot power. If you don't want that then don't use power saving.
The kernel configuration docs ("conservative" frequency scaling governor) have this to say about AMD64 CPUs:
"If you have a desktop machine then you should really be considering the 'ondemand' governor instead, however if you are using a laptop, PDA or even an AMD64 based computer (due to the unacceptable step-by-step latency issues between the minimum and maximum frequency transitions in the CPU) you will probably want to use this governor."
The governor is already set to ondemand.
Only Intel's advanced C-steps affect perfomance.
Speedstep's perfomance loss is around 1%.
The point those docs are trying to make is that "conservative" might be a better choice than "ondemand" and that AMD's freq scaling is much slower than speedstep (though it fails to mention on what chips it's actually slower; Athlon? Phenom? Phenom II?)