I couldn't find the article that states that light penetration varies with the color of the channel. Not sure that is the case. Even if it is true, it is a matter of degree.
Yes, that is the case. Silicon (not silicone) is transparent to IR, red penetrates deeply, green not so much, blue not well at all. When I was in grad school, I had a professor (who earned his PhD in Carver Mead's lab) who encouraged me to investigate and design a similar imager as a course project, which I did. As I was using a commodity CMOS process, I was limited to a 2-layer stack instead of 3. Sadly, my lab partner was unwilling to commit to the testing of the IC which would have to continue on into the next semester, so we never fabbed the chip.
Using the penetration depth vs. wavelength characteristics of silicon means less control over the spectral width of the bandpasses (varying the layer thicknesses), and more importantly the steepness of the cutoffs. The penetration depth is a statistical distribution, not a "435 nm light penetrates exactly 0.35 micron, no more no less" exact value. I'm guessing that there's more to RAW conversion with the Foveon than just separating the colors; there's probably some post processing to better separate the primaries, but that's just a guess based on my planning for what I would have had to do for my 2-color version.
Sadly, these days I'm not involved in IC fab and this design project was getting on 10 years ago. Some of the specifics have faded, and of course the X3 is almost certainly more sophisticated than my design and with a process that can be tuned to their specific needs. However, I'm pretty sure ( insert winky smiley here) that there aren't color filters between the layers as in color film, and so randomizing which color is on top simply isn't an option using the Foveon process. For that matter, there really isn't going to be a silicon-based process that picks up blue photons at the bottom of a 3-pixel stack. Even without any color filter, those photons simply won't penetrate that deep.
And a thin silicon layer on top won't pick up many red photons, so the detection efficiency will be terrible.
