Single-ended data transmission, as used in RS-232, is limited in noise immunity, so higher-speed data is commonly transmitted with differential signaling. Unfortunately, this requires twice as much wire.
AC power transmission faces a similar problem: unless you’re willing to use the planet as your current return path (commonplace in parts of Brazil and for submarine cables) you need to carry the current from the power plant to the load — and then back again. In AC power transmission, this problem (as well as the problem of starting large AC motors) is solved with three-phase power transmission: three wires carry alternating voltages 120° out of phase with one another. The sum of the three phasors is zero: always approximately, and for balanced loads like a three-phase motor, exactly. This permits carrying thrice as much power over only 50% more wires than the single-phase approach used in household outlets.
If you replace your twisted-pair data transmission lines with twisted triads, for 50% more wires you can’t get thrice the data rate, but you can get twice the data rate, with roughly the same noise-immunity benefit you get from twisted-pair differential signaling. The third wire carries a balancing voltage and current which maintains the triad’s net voltage and current at 0; the mutual inductances and capacitances of the three wires remain constant along their lengths, just like in a twisted pair, and because these mutual inductances and capacitances are are all equal, crosstalk can be accurately canceled. At low speeds this balancing voltage can be arranged straightforwardly by driving the third wire with an opamp that maintains a constant reference voltage at a summing Y junction; higher speeds might require more elaborate drivers that multiplex the third wire among, in the bilevel case, four compensation voltages.
Attempts to extend this approach to differential quads, quintuplets, and so on will suffer from the fact that the balancing wire and signal wires cannot all be equidistant, at least in three-dimensional space. The result is that either data edges on one signal wire that couple equally to two other signal wires cannot be perfectly canceled on both of them by a single balancing wire, or data edges on one signal wire that couple unequally to two other signal wires cannot be perfectly canceled on both of them by a single balancing wire. Also, if the voltage levels on the three or more signal wires were uncorrelated, the balancing wire would be subject to much larger voltage swings. Still, MIMO approaches could yield fruit with n wires while retaining net zero voltage and current on the wire bundle; in theory it’s just a question of solving a well-conditioned, very nearly linear system of n equations in n unknowns, one of which is the common-mode noise.
The problem of unequal voltage swings and complex analog line drivers can be solved, at the cost of reduced data rates, with a constant-weight m-of-n code like the 2-of-5 code. In a sense the 1-of-3 encoding SETUN used for its three-way flip-flop(-flap?)s was a constant-weight encoding providing 1.58 bits over three “differential” wires. Decoding such a constant-weight code could be considerably simpler than a more sophisticated MIMO approach, merely amounting to comparing each wire to a threshold from a summing junction; a balanced code on 8 wires would provide 6.1 bits of data per clock (70 valid codes), and a 9-of-19 code on 19 wires would provide 15.9 bits per clock (8314020 valid codes).
More generally, if you can statically estimate (linear, memoryless) crosstalk coefficients between particular wires, you can build a summing junction that computes the “predicted” crosstalk for each wire, to be used as the comparison threshold for that wire, thus mostly canceling that crosstalk.