It sounds like you could maybe make substantial progress by just separating the differential pairs (?) by a millimeter or two of dielectric, giving you a ribbon cable, with much lower crosstalk than the round kind. Bonus points if you color the dielectric rainbow colors.

I feel like it's harder than shielding and tolerances, given that longer passive cables don't seem to exist despite the very high prices people are paying for active cables.

There might be issues of attenuation; we're talking about signals in the GHz range, where you have to use waveguides instead of wires to get low losses.