Distortion:
Sure, why not. Or, more generally, if not distortion per se, then non-LTI behavior. For example, thermal expansion of the core/winding will not cause distortion as such (i.e., on a per-cycle basis), but the change in value will eventually change the signal amplitude/phase, and therefore it is not a time-invariant system (i.e., the inductance parameter varies over time, in this case due to temperature).
This will be a ~ppm scale effect, at best, for the most part. The core, if present, will distort significantly more than this.
Mechanical effects are significant in some types of ferrite, which exhibit relatively high magnetostriction (strain (change in size) resulting from magnetic field magnitude). In this case, per-cycle effects are perceptible, and electro-mechanical coupling is seen. Much like a speaker or piezo, mechanical resonances manifest as peaks and dips in the electrical impedance.
Regarding cross section, rectangular is probably more popular because of higher power density. Also, increased core permeability makes the wire conductivity much less important. And for almost all materials, core loss dominates*, anyway.
There aren't any EMI/radiation or near field problems with an odd cross section, as long as it is still a solid of revolution. The toroid core depends on its rotational symmetry, not on the shape of the cross section (but yes, winding resistance will be least for the circular case).
*In a real power inductor application, ripple might be a modest fraction of DC bias, so that total copper losses can be comparable. The difference is due to DC heating. The AC heating is still dominated by the core.
For an uncored inductor, a square winding path is of course worse. Curious about that paper, that they went to the trouble of fabricating these things, but they made the one mistake that would've won them another factor of 2: rounded conductor cross section with even spacing of turns.
This is easy to demonstrate with a
helical winding calculator: set pitch = 2 * wire diameter and observe the Q. Then squash it together, say pitch = 1.1 * wire diameter, and observe again. (The inductance is higher in this case, but the AC resistance is MUCH higher, and the Q lower.)
They also claim to cancel out the solenoidal field from a single layer progressive toroid winding, by stacking two oppositely. Uh, right. That makes a quadrupole, which drops off faster at a distance but is most definitely not a cancellation!
The other links aren't very substantial. Transformer winders follow drawings; they don't need much design expertise in house, let alone theoretical knowledge. No proof is given, only claims made.
Tim