Faster TT - Attack Hills or Kill Descents?

In discussing time trial strategies with other riders, I've found that everyone seems to have a different approach to dosing their power over a rolling TT course. There's the 'don't start too hard' rule, which I think is very valid -- I make that mistake often. Then there's the question of when to stand out of the bars, etc.

But the question that I was most uncertain about was whether to put more effort into climbing or descending (assuming that your average wattage for the entire ride is constant). A very good time trialist I know says he powers the descents and that gives him an advantage. I've always heard that I should put out more effort on the climbs and recover on the descents. One coach told me to stand on the last 25% of each roller to maintain momentum. So I've done the math.

I modeled the situation using a spreadsheet allowing input of temperature, elevation, power, frontal area, friction coefficient, rolling resistance, and weight for various course sections. I constructed a 1-mile TT course with a 1/2-mile, 6% climb followed by a 1/2-mile 6% descent. I used aero coefficients that represented climbing on the hoods and descending in the drops. The average power over the course is 250 watts for all three situations.

Constant effort:
If I generate a constant 250 watts over the course, I get the following times: climb (2:58)+ descent (0:45)=total (3:43).

Harder climbing:
If I generate 270w for the climb and 178w for the descent, I get: climb (2:48)+ descent (0:46)= total (3:34). That's 9 seconds faster than the constant power ride.

Harder descending:
If I generate 240w for the climb and 290w for the descent, I get: climb (3:05)+ descent (0:44)= total (3:48). That's 5 seconds slower than the constant power ride.

I tried the same thing for a 3% climb and got similar results. The end result is this: If you are going to ride with a non-constant power output, it's better to hit the hills hard and take it easier on the descents. Extra power output doesn't do you as much good when you are at higher speeds because it gets eaten up by wind friction at a 3rd power rate.

Why does my friend the good time trialist think it helps to kill the descents? Probably because he kills the descents but continues to kill the climbs too! So he ends up with a greater average power for the ride. The moral of the story is "If you have to recover somewhere on the ride, do it on the descents, not the climbs."