You can always trick the numbers by implicitly requiring a super-human power output when climbing, but at constant power for the sake of comparison the additional drag at the higher downhill speed will kill the average (you won't translate all the potential energy gained by climbing into velocity). In a vacuum that wouldn't be the case so your average would only be lowered by the second order mechanical power losses on the way "down".