First VeloNews. Both Cervelo S and Felt AR frames were tested by both VeloNews and Scott. VeloNews tested the Masi Competizione, which they claim is similar to a Trek Madone. Scott tested the Trek Madone, so I include the results from these tests as well. VeloNews' tests were with Zipp 404 wheels except for the Masi which was tested with Fulcrums. However, I adjusted the VeloNews data based on the (considerable) improvement the Zipps had on the Ridley and Cervelo in their test, which were each tested with both the Fulcrums and the Zipps.
VeloNews tested at positive and negative yaw, while Scott reported only wind from one side of the bike, so I averaged positive and negative data from VeloNews. In each case, both for the North Carolina windtunnel used in the VeloNews test, and in the Mercedes windtunnel used by Scott, I assumed an air density of 1.2 kg/m³ to convert from force (VeloNews) or power (Scott) to CdA (the coefficient-of-drag-normalzed cross-sectional area for wind resistance).
The Scott test was "bike only with similar wheelset". VeloNews tested complete "stock" bikes with both "stock" and Zipp 404 wheels.
Here's the comparison:
First, the Scott numbers are obviously lower. Other than that, the absolute difference between the Felt and the control bike is rather similar in the two cases. The difference between the Cervelo and the control bike is larger in the VeloNews test. As a result, the Cervelo does better than the Felt in the VeloNews test, while the two are much closer in the Scott test.
Curious is the shape of the Madone curve versus the aero bikes. The Madone has a steeper CdA-versus-yaw slope. In the VeloNews test, all bikes, after adjusting for the same wheels, had similar slopes. I wonder if the Scott test was done with Firecrest wheels on the Madone, or like in the VeloNews test, there is an undisclosed exception made for the "control" bikes. The Scott test had three control bikes, and all three followed similar slopes, while all three aero frames followed a reduced slope.
One issue with windtunnel tests is you need to subtract off the "tare": the wind resistance due to the set-up and not due to the bikes. So it's possible the tare subtraction differs in the two tests. I wouldn't necessarily expect the two tunnels to yield the same result for the exact same bike. Then there's the issue of air density. I don't know what the relative air density is is in the two tunnels. But there's countless other factors which could differ in the two tests, starting with frame size. I simply don't know much without photos.
Then there's the Tour Magazine tests. These were done with a dummy mounted on the bike. Results with the dummy are more variable than those without the dummy, at least to my eye. Here is a comparison of the Felt, Cervelo, and Cannondale tested by Tour to the Felt, Cervelo, "round-tube" control, and Scott Foil tested by Scott.
The difference here is just huge. The Scott numbers, around a CdA of 0.22, would be about the best you get from a time trialist with a super-optimized position like Levi Leipheimer or Dave Zabriskie. The Tour values, around 0.31, are much more typical of a mass-start position. Tour showed a photograph of their dummy set-up, while Scott did not. Maybe the Scott dummy had no arms. That would be my guess. Such a simplification would facilitate set-up.
Looking at the Tour result, that low-yaw Cervelo result is just weird. I see nothing so anomalous in the Scott test.
In the Scott case, the Scott suddenly does substantially better than the Cervelo and Felt with the dummy on the bike. Why? Looking at images of the bikes, I can't see a good reason for this. You might think the Cervelo and Felt have some aerodynamic feature, like an aero post, which helps them much more without a rider than with the dummy mounted. But all three bikes have aerodynamic seat posts.
What do I conclude from all of this? Basically I think it's clear that the aero frames do improve things, and significantly, albeit less than other factors like body position (or perhaps if the rider has arms or not). It also appears the Scott is a legitimate aero frame, in the same class as the Felt and Cervelo, despite being much lighter. In the Tour test other so-called aero frames, like the Canyon and Stevens, did not do so well in the wind tunnel relative to their Cannondale control.
But it's obvious wind tunnel tests are hard to do. Make different assumptions, get a different result. Looking at end-results only, without details about the assumptions and the set-up, is of limited value. This is especially true when you test with a dummy rider, which in principle seems the right thing to do, but then you need to make sure you're meticulous in setting up the dummy in exactly the same position, and that fitting the dummy to the bike in this position doesn't introduce any biases for one frame versus the other since different frames fit different riders differently (especially a dummy whose position can't be tweaked even a little without confounding tests). Hats off to Scott for showing results both with and without the dummy rider. They could have shown only the dummy results, and claimed unambiguous superiority.
added 13 Jun 2011: after writing this I learned Scott tested the bikes with two water bottles and cages, while the other tests were done without water bottles. This provides an advantage to Scott (as it would to Litespeed, for example) for a better interface between the truncated down tubes and the bottle shape. Second, it would award more optimized placement of the water bottle holes. Bottles may either increase or decrease wind resistance relative to no bottles, but the much lower CdA's apparently evident in the Scott test relative to the Tour test, for example, still puzzles me.