The spectral quantum efficiency of the typical sensor well tends to peak at about 700nm and fall away toward blue and red.
Red has the lowest QE because the longer wavelength lower energy photons go deeper into the depletion region and may not end up being stored as electrons.
Also red photons may end up in adjacent wells.
This is explained by Holst and Lomheim in "cmos/ccd Sensors and Camera Systems"
To check this effect with the Pentax K-01 , I took photos of a gray card and a red T shirt, and directly converted (before the Bayer de matrix and de-gamma) the raw file to a hex file.
This jpg is a copy of the spreadsheet I made to show and quantify the raw pixel values.
https://app.box.com/s/7z4i3w4iey5h7r2eqmblgs28nduimcf5
Note that on the gray card LAB 50,0,0 , exposed normally so the camera histogram indicated about 50% brightness, the red wells are only 5.7% of full capacity.
In decibels, the red wells are typically at minus 24 dB ( or minus 4 EV stops) of well capacity, and less than half of the values in the green wells.
To prove these levels, I took an overexposed photo of the bright blue sky, to show that all wells do reach full capacity (approaching FFFF on the 12 bit K-01)
It is apparent that the histogram displayed on the camera lcd is for the resultant jpg, after application of the Bayer de matrix coefficients, and the gamma etc.
The camera histogram does not give direct information about the utilization of the well dynamic range.
This low red level has a worse (by about 20dB compared to full well) signal to noise ratio,
and worse (by 16 times), quantization noise by the adc because the adc on red channel is effectively only with a resolution of 8 bits (1 part in 256)
All this "truth" might sound more depressing than it is. I don't know if the pundits consider this in the performance tests.
However the Bayer sensor works rather well, I think. I suppose it is due to our forgiving eyes.