Yeah -- it's a geometric series. If the slope were 1/1, the series would terminate with just one pole, because that's what a capacitor or inductor is. The slope is 1/2, so the R and (L or C) values increase (or decrease) with a constant factor per stage. The number of stages used is determined by the approximation error and bandwidth. (If the slope were 0, it can be approximated by two things: an ideal resistance, or an infinite series of equal L and C -- a lumped equivalent transmission line. Math is fun like that.)
AoE3 shows this in the circuit for pink noise generation, including a bandwidth and error plot.
I've also done same as the OP on a previous occasion:
The capacitances can also be roughly expected, as the laminations have considerable capacitance between them, which is reflected as capacitance at the terminals. In the same frequency range, the winding's own self-capacitance and leakage will become relevant, as well as complex wave effects; this, and laziness, are the reasons for the poor curve fit at high frequency.
You can also write a true diffusion component in SPICE: a frequency dependent resistance. (Coilcraft uses this element in their SPICE models.) Unfortunately, this has no concise time-domain representation (the Fourier transform is a sqrt function over all time, so SPICE can only solve it with a convolution, which has to be evaluated per timestep, and the convolution vector is as long as your simulation is...). In transient simulation, most simulators either ignore it, or go really slowly and give terrible numerical errors.
Tim