Ah, well then, the problem is you're probably looking too close in. EMI problems can only be understood locally, in specific circumstances. Namely, when the block in question can be represented as an N-port...whatever that is, right? Otherwise, it must be understood holistically -- system wide.
Which is where that advice is coming from: treating a three-terminal regulator as such. If it's four-terminal, it's actually three ports and ground, or four ports against whatever the nearest ground-like reference is. Fewer ports is easier to reason about, easier to treat. That's why you always want to keep loops tight, and grounds nearby and contiguous: you're reducing the number of ports you need to evaluate.
An ideal port, is simply a winding on an ideal transformer. No common mode current: any current entering one pin, leaves the other, period. A port is two pins, there is only one voltage and one current, that's it.
A port connects to a block or network or system, within which, undefined things happen, then the signal maybe shows up at some other ports of the network, in which case you can measure the gain and such.
So of course, a three-terminal device, with common ground, can only be a two-port -- and with those ports defined with a common ground, there cannot be any common mode! So the problem reduces to a simple DM filtering problem, and you only have two ports to filter. Great, right?
In reality, we don't have ideal ports. We always have a common mode impedance. Real transformers are made with pairs of conductors: otherwise known as capacitors. (Or more generally, transmission lines, which is to say, pairs of conductors with a well-defined length.)
So, we might have to model a 4-terminal power supply as three ports with a common ground, or four ports against a global ground plane. We can build test fixtures this way, where each port comes up to the EUT in a 50 ohm coax cable, terminating its shield to the ground plane, and the signal wire going up to the EUT. We connect whatever sources and loads we need to the far ends of those cables, usually through filters so we can measure what the EUT is doing at AC frequencies.
(When would we use N ports for N terminals? When the EUT is badly behaved. If it's a bare PCB for example (or a PCB inside a nonconductive enclosure, doesn't matter), relatively sizable compared to the frequencies we're dealing with, then that PCB area can radiate in and of itself. Then we would have an unaccounted-for common mode. Whereas if it's in a grounded enclosure, we can take that enclosure as reference plane, tying one terminal off to it (if this is feasible, and representative -- yes, still more conditions apply), and treat it as N-1 ports.)
I'm guessing, you're going to have something like, a battery which is two short wires hanging off a connector; a power supply module; a, whatever the core of the thing is, amplifier, processor, whatever; and probably further inputs and outputs to that. Maybe the PS is on the core board, that's fine.
Well, between the PS and battery, you don't want to have unbalanced (CM) AC going to the battery, which is a big hunk of metal, also known as an antenna. If we think of the two wires as two ports against ground, we can figure, if they're moving oppositely (differential mode), the currents and voltages cancel and no radiation takes place. Or maybe a little, from the loop antenna formed by the wires themselves -- but we can twist those wires to keep that loop small, so it's not too bad. If we think of the two wires acting in parallel, as if shorted together at AC -- which is usually the case as power supplies generally need bypass capacitors at the input and output -- then if the PS is generating a CM voltage, the wire and battery can radiate like a monopole antenna (or if there's not much connected to the core PCB, then as a dipole between those two parts, battery and PCB).
Continue this thought process, recursively for every connection on every board, and you'll develop an equivalent common mode circuit for the whole system. This will be your basis for further reasoning -- where to put filtering, if it needs to be CM or DM or what.
Tim