I haven't heard of the mod, thanks for the link.
The problem they fix there is however as far as I understand only image smearing (which is important too of course). Anyway, the color filter array sits closest to the sensor surface, under both cover glass and microlenses.
The problem I'd like to see solved is crosstalk, the issue when light comes in through one color filter on the sensor surface (say a red one) and then inside the chip jumps over to a neighboring pixel (say a green) and then mixes up colors. This can be seen as desaturation and in severe cases mazing. A thinner cover glass won't help for that. In the past sensors this happened a lot as the photo diodes sits deep down in the chip with open space between neighbors where light could freely pass over to the neighbor if coming in at an angle. Kodak CCDs had put in walls (light shields) between pixels, but those where not available in Dalsa CCDs and not in Sony CMOS.
In theory on a BSI sensor the photo diode sits directly on the surface of the chip, virtually sandwiched to the color filter. This means that light has no chance to skip over to the next pixel. If practice match theory there would be virtually no crosstalk with this sensor. Color cast is probably still there though as I think part of that is due to changed color filter response when light comes in at an angle. Color cast is correctable though, while crosstalk really isn't.