The light sensitivity of a AF point's line sensor, ignoring filters etc., is mainly governed by the size, i.e. light receiving area of its pixels - the bigger the pixels, the better the sensitivity, but the bigger the pixels the lower the inherent resolution of the AF line sensor.
Phase Detection AF is based on images taken from opposite sides of the camera main lens (exit pupil) projected by the Secondary Image Reforming optics onto said line sensors. With the subject in focus these images will be projected with a certain distance onto the line sensors, a distance referred to as "baseline". In the OOF state the image projections will be either closer together (front focus) or farther apart (rear focus) and hence the deviation from the "baseline" yield information about the amount of defocus as well as the direction.
That a line sensor is specified as F5.6 or F2.8 has to do with it's baseline, where a F2.8 sensor has approx the double baseline of a F5.6 line sensor. Back projection of the line sensors onto the main lens's exit pupil hence means that F2.8 line sensors get their images at a greater distance from the lens optical axis than do the F5.6 sensors. A F5.6 sensor will hence always be able to use a max aperture F2.8 lens whereas the opposite potentially would have the F2.8 line sensor pair projected outside exit pupil aperture of a lens with max aperture of F5.6..
The main enemy of PD-AF is vignetting, and even though the "back projection" of a F2.8line sensor might actually fit inside the exit pupil aperture of a lens having max aperture of say F4.0, using the sensor in this context might lead to vignetting of the "outer regions" of the images projected onto the line sensors, resulting in front focus.
Looking on vignetting data, you'll notice that most large aperture lenses are significantly vignetted wide open, one reason that say F1.4 baseline line sensors are not used. But large aperture lenses OTOH also mean more light, which works in the same way for AF sensors as for image sensors, i.e. faster exposure and less noise at a given shutter speed and light level.
To work out the projected image distance on the AF line sensors a certain amount of contrast (i.e. difference between the brightest and darkest part of the image seen by the line sensor) is needed, which can lead to problems with both too bright and too dark scenarios. To that comes that the latter scenario also has the problem that image noise will be added to the pixel data resulting in imprecise image distance determination (focus error)
Thus as such F5.6 line sensors are not more light sensitive than F2.8 sensors - if they use the same pixel size, but F2.8 sensors with their larger baseline have a better resolution in absolute terms, at the expense of a reduced OOF range handling capability (assuming the same line sensor length).
Along those lines, cross sensors are not more light sensitive than uni-directional sensors per se, but having the double sensor area available and the capability of reading contrast in more directions the chances of finding a sufficiently contrasty pattern at low light levels should be higher with cross point sensors.
How the 1/3 DOF accuracy with in the High Precision mode available with F2.8 AF points and F2.8 lenses or brighter is achieved , is a Canon secret. A qualified guess would however be, that since AF resolution as already calculated below the AF pixel pitch using linear or polynomial regression, the precision of this technique is enhanced by having more light available (read: higher contrast and better signal to noise ratio)...
~ hans ~