So. I [re]read wikipedia and the hyper-physics site. Wikipedia in particular has a nice reference to a paper by Kenneth Cartwright, which is very approachable. In summary, there are three resonant points.
f_0: series resonant point (X_C = X_L); works for parallel given low ESR
f_p: (phase = 0) = (power factor = 1)
f_m: max impedance
From the Cartwright paper I was easily able to construct a spreadsheet that allows me to calculate either f_m or C (given the other), for a given L+ESR. The result is in very good agreement with Circuitlab and LTSpice. And I can see both the f_p and f_m points on the plots. Now I have a good enough grasp of this to properly consider the effect of R_s.
But I'm still at a loss as to why the minimum current does not occur at the f_m maximum impedance frequency. It also is not at the f_p frequency. Cartwright does not consider this at all; in his circuit he uses a 1A constant current source, which at least one good reason to do so is to remove the effect of R_s!
Why doesn't V=IZ apply? Z is not plotted, V is plotted. It occurs at f_m, max impedance. Ergo, given V=IZ, V is at a maximum, Z is at a maximum, shouldn't I be at a minimum at this same frequency?
EDIT: Possible answer: V is not the total circuit voltage, it's the voltage at the divider point. As Z (of L//C) changes, R_s remains constant. This means the division ratio changes with frequency. The shift in I_min away from f_m reflects that V is not changing at the same rate as Z!
Re-attaching a couple of relevant plots from a previous post.