This page covers more specific and technical topics than the FAQ section, but for the reader's convenience is organized for the moment like a FAQ list.

A common misunderstanding is to think of the mutual resistance term as a dielectric loss. However, the dielectric loss should be more correctly modeled by a off-diagonal real term in the capacitance matrix (anyway, FastCap does not compute this conductance term, as in the FastCap equations the dielectric is assumed lossless) and FastHenry does not consider any dielectric. So where this mutual resistance arise? A mutual resistance term means that, in response to a variation in the current flowing in one segment, there should be a variation in the voltage across one other segment. This is a direct consequence of the proximity effect: at high frequencies (when the physical dimensions of the structure are in the order of one wavelength) the current is no more uniformly distributed along the conductors cross section; at these frequencies, a variation in the current flow in a nearby segment causes a variation in the effective cross section available for the conduction in the given segment. This results in a change in the resistivity of the segment caused by a current flow variation in the nearby segment: that is, a mutual resistance term.
There is also another situation that can cause this off-diagonals real terms to appear, even in stationary condition. This is the case of two current loops with a part of the path in common (e.g. two lines and a common ground plane through which the current loops close). But this case is trivial and arises only because of superposition of the effects: a variation of the current flowing in one path causes a variation also in the voltage across the common part and this reflects on the total voltage applied to the second path.

Very often there is the necessity to shield in some way one conductor from one another to reduce inductance. Unfortunately, inductance coupling does not work like capacitive coupling, for which a grounded line between two signal ones makes the trick; or at least not entirely. Let's think about how things work at low frequency, for a start. We have static fields. The electric field does not penetrate into conductors, because if we had an electric field inside, the free charges into the conductor would start to move towards the surface, so there wouldn't be equilibrium. Therefore, due to the accumulation of charges on the surface, we have a shielding effect for the electric field.
Well, the same is NOT true for magnetic fields, if we are considering materials with relative permettivity (mr) near one (that is, non (ferro)magnetic materials). In this case, the field penetrates the conductors and is almost not disturbed by them. The result being, you cannot shield two conductors using a third one in the space between. However, when the frequency increases, things start to change, because of the proximity effect, that is, eddy currents start to be induced on nearby conductors. These induced currents do have the effect of shielding one conductor from the other; however, one single wire between other two hardly produces any difference. A ground plane would provide a more marked shielding effect (though still very small).
FastHenry could be used to perform a couple of experiments to get in touch with some quantitative results: first of all, simulate a wire alone in the space, from 1KHz to (say) 10GHz. Then put a ground plane somewhere near the wire. It's not needed to create a port for the plane to see its effect, so leave only the port for the wire in the simulation; this helps to better understand what's happening. Running FastHenry, we can notice that at low frequency the results are the same than before the insertion of the ground plane. However, when the frequency increases, also the high-frequency effects are more pronunced: that is, resistance rises, inductance drops more than before. The secon experiment consists in simulating two wires in free space and then inserting a third one between the first two (again, not creating a port for this last one). By comparing the results at high freq, we could notice some very slight differences in the inductive coupling (off-diagonal) terms. Using a ground plane in between, we could get some more shielding, but this solution is very often not feasible; moreover, remember that this shielding effect not only is small but also is visible only at high frequencies.
However, if it is true that the magnetic fields are not shielded by materials with (almost) unitary mr, it is still possible to reduce the loop impedance of a conductor. The loop impedance depends on where the 'return' path for the current is located; therefore, if you insert a third conductor between the two conductors you are trying to shield from each other and connect it at both ends to a ground plane, a current can flow to reduce the magnetic field. But since you must be able to make current actually flow in your conductors, you cannot appreciate this effect completly from the inductance matrix but it is better seen with a circuit simulator. This concept, however, is very delicate, since shielding depends on where the foreseen 'return' path for the current flowing in your lines is (i.e. in a nearby gnd plane, as could happen in a PCB, or very far on some other conductor, as could happen in a dense connector).

[Many thanks go to Stefan Weber of Fraunhofer-Institut für Zuverlässigkeit und Mikrointegration (IZM) for his comments on this topic and the useful discussions we had.]