Iliah Borg has pointed out that he does his UniWP calibration in a monitor space with native primaries and a gamma of one. Not using the native primaries might be one source of inter-channel crosstalk. To find out if this is significant on my NEC PA301W monitor with the D800E, I made three exposures at the corners of the native monitor space and recorded the average values of the three camera primaries in the raw files.
R=255, G=0, B=0:
R=0, G=255, B=0:
R=0, G=0, B=255:
There is still significant crosstalk. Comparing the two matrices one over the other, with each primary normalized to one, we can see they are quite similar: