The notion that the emission is roughly spherical is (IMHO) “wrong”.
It is (should be) omnidirectional as a toroid in the same plane of the device, and likely the performance when front-facing is due to either the “thickness” of the toroid or to reflections (or *whatever").
When it is placed in the same orientation as it should be (fixed to a ceiling) it is very likely that the signal is better if the client is at a lesser height.
A picture of the internal of a cap ac2, the antennas are on the PCB:
A picture of the internals of a cap ax2, the antennas are external:
Is the (aluminum? inox?) ring an antenna (front facing) or a shield? (I believe the latter but cannot really say).
Since in all cases you tested the Ax has better signal levels than the Ac, the differences in speed are more likely to be connected to some quirks in software, drivers (or whatever), the configurations (right or wrong as they may be) are the same, so they are comparable, unless there is some differences in the way the commands are implemented in wifi-qcom-ac vs. wifi-qcom.