Ah...I finally found it. (the book)
A wratten 29 (red) at its primary wavelengths of between 620-700nn is passing about 90% (90.5% at 700nn) of the red light-meaning about 90% efficient. A green 61 at it's peak transmission of 520nn is about 40% (which is one factor of why there are usually two green collection sites) but the blue 47 at peak of 440nn is 50%. The fact that there are two green sites in a Bayer array and a 90% transission efficiency of the red and you see that if you only have 1/2 of the blue light going through a 50% efficient filter and only one site to collect from, the loss of blue data from a digital capture will result in a metameric failure when comparing a capture done under tungsten and one done at D65.
Which is another reason that white balancing using a two axis slider can be problimatic at best. There just isn't a lot of blue data captured with a digital sensor.
We do not really need to consider Wratten filters to answer the question, since the answer is in the raw files. In some experiments with my D70 I photographed a MacBeth Color Checker under 40 watt tungsten illumination (2950 K according to ACR) and daylight and examined the raw files (converted with Iris and with the pixel values multiplied by 8 to convert from the raw 0..4095 to the PS 0..32767 "16 bit" format). For convenience, the results are displayed in Photoshop as 8 bit. Since the sensor is linear, the pixel value is proportional to the luminance seen by the sensor elements.
Here are the results for daylight, showing the pixel values and histogram for the white patch:
Here is the same for tungsten, ~2950K
To achieve white balance, one multiplies the red and blue pixel values by a scaling factor as shown in this [a href=\"http://www.pochtar.com/NikonWhiteBalanceCoeffs.htm]table[/url].
For 2950 the RGB multipliers are 1.32, 1.0, and 2.9. In terms of f/stops, the red channel is down 0.4 stops and the blue channel 1.54 stops.
For daylight the RGB multipliers are 2.06, 1.0, and 1.6, or expressed in f/stops the red and blue channels are down 1.04 and 0.68 stops respectively. These factors have nothing to do with the fact that the Bayer array has twice as many green sensors as blue and red sensors: the missing colors is each Bayer pixel are filled in by interpolation during the demosaicing process. I don't think the differences have anything to do with metamerism.
The from the above pixel values, the ratio of light falling on the blue sensor in daylight to that with tungsten 2950K (daylight:tungsten) is 1.68 or about 0.75 f/stops. This is about the same change in ratio that takes place when you change the ISO on the camera from 100 to 160, and this does not affect performance that much. At low ISO we get very good results with the digital sensor at 2950K. The blue sensors have plenty of light to work with, and they are less taxed in this situation, than when base ISO is increased from 100 to 200 with daylight illumination.