I'm not sure I understand the EZNEC results. I modeled the antenna over mininec ground. The result was -0.04dBi at 23 deg TOA. The current in each of the radials was 0.5 amps, half of the supplied current. I then added a wire from the junction of the two radials to mininec ground to simulate a common-mode path. The result was 0.0 dBi at 22 deg TOA. There was 0.15 amp in the simulated path to ground and 0.57 amp in each of the radials. The simulated common-mode path to ground increased the current in each of the radials by 14%.
It's well known, and has been publicized for a long time, that a few elevated radials are problematic for common mode. Also, any ground path to earth increases loss.
Even a groundplane with four 1/4 wave radials requires a common mode isolation method or the antenna will be sensitive to feedline and mast lengths, and ground path changes.
There is some optimal ground path impedance that presents the most loss. A mininec ground may or may not allow that impedance.