This is a little bit confusing part. I clearly understand that in ethernet cable fields are coupled between wires of differential pair. On PCB on IC/PHY side it is also known that differential pair fields are coupled to ground (reference) plane - coming from that, that on PCB diff. pair is actually just two single ended transmission lines close by. And then there is this confusion land where cable connects to PCB and goes to transformer... To what are the diff. pair lanes coupled to ? To ground, but why would they, as magnetics on both ends of cable give then isolation. Or do they still couple to each other ? If so, then every Zdiff. calculation tool is unusable as those assume some reference plane.
The whole point of using a differential pair for signalling is that there is no need for a reference plane to compute the odd mode (the signalling) impedance. When you do the calculation, there is an implicit reference plane mid-way between the two conductors. As the conductors are driven with opposing voltages, the voltage mid-way between them is always 0.
You can also define an even mode impedance, which is the same signal being carried on both conductors, and this is with respect to some remote reference plane. In the case of ethernet, this even mode signal is poorly driven and received, as it has to couple capacitively across the isolating magnetics at either end. This is the signal responsible for EMI via conducted emissions, so on many cables it's attenuated by lumps of ferrite on the cable that raise the even mode impedance further.
The PHY-side Ethernet wiring has both intra-pair differential and common-mode impedance coupling to ground (generally, 100 and 50 ohms, respectively.)
The field-wire side of Ethernet only has intra-pair differential coupling (100 ohms) and no common-mode impedance to any reference (more on that below.) The field-wire cable in fact has no DC path to ground at all, and can be any voltage with respect to ground, even a potentially damaging one.
(Power-over-Ethernet exploits Ethernet's field-wire isolation by inserting DC offsets on the field wires to transmit power. That's another discussion, but illustrates the point.)
In practice, the Ethernet magnetics has the field-side pairs terminated using what's called a Bob Smith termination: each field wire ties to a 75 ohm resistor, with a common tie to a high-voltage cap to ground. This terminates the field wires into a 75 ohm single-ended impedance to a 'virtual ground', with the added benefit of reducing the possibility of DC buildup.
The magnetics also includes common-mode filtering which further improves rejection of any transients that couple onto the field wires. A robust design will also include ESD protection for the PHY-side pairs to further reduce the effects of any transients that make their way through the magnetics.