Actually in some heavy FP algorithms it would be more then twice as fast as llano per core (when AVX is used of cause).
I think many media encoding software (and other calculation-intensive software) will be ready with AVX-optimized versions for the SB launch.