Do you know of a NEC article that addresses and regulates this case specifically? We had a lively discussion about it over on the Photovoltaic forum a while back but I don't think any language in the code was found to say that explicitly.
In a related story, I saw some photos the other day of a service that had 4 sets of three phase 208 going into a seven terminal Polaris block, and one of the seven terminals fed another four terminal Polaris block. They had 4 taps to service discos - two from the seven terminal block and two from the four terminal block. Is this sort of thing normal?
There is a problem with splicing just one set, and not splicing the others, because of the issue of "be of the same length/terminated in the same manner". Splice just one of the conductors in parallel, and you need to splice all. This doesn't mean that you have to splice all the remaining phases, just the phase in particular that has the spliced conductor.
Tapping another circuit onto a mid-point of a conductor, with an insulation piercing connector or similar, is a blindspot of the NEC on parallel conductors. There is nothing in the wording of the code to explicitly prohibit a mid-point tap onto just one set of the conductors. The reason is that technically you didn't terminate the run conductor at that connector, so it is difficult to say that you didn't terminate all sets in the same manner.
As I see it, do not break the symmetry by doing anything intentionally. I can show you with a Kirchhoff law calculation, with a mid-point tap onto one of the conductors in parallel, and a result with a currrent imbalance that exceeds the ampacity of the conductors.
