I see that the modified Wilkinson behaves equivalent to regular type in a restricted frequency range (< 2 Fc), but only if the common mode impedance of the flipped transmission line doesn't disturb too much.
As previously explained, the basic SPICE (or Genesys) transmission line doesn't model common mode impedance, it always behaves ideal. You can supplement the simulation by adding a second TL component for the common mode signal (coaxial sheet to vicinity). The basic problem is that the electrical length of the common mode line will be usually shorter than the coax line with dielectric, thus it's not λ/4 in length and the shunt impedance won't be eliminated.
HFSS can be expected to model the transmission lines completely, if the result should comply with the real circuit if all parameters are modelled realistically.