HFSS - Antenna directivity with infinite ground plane

Hi guys,

I'm using HFSS to simulate an antenna, and I've run into an interesting problem.

The antenna is surrounded by a standard airbox, lambda/8 at the longest wavelength away from the antenna, as recommended.
I'm interested in simulating the effect of an infinite ground plane on the antenna, so I use a perfect E boundary on the bottom face of the airbox, and I use radiation boundary on the others.

Now, in the perfect E boundary settings, I can choose "infinite ground plane" which is what I need. HFSS help tells that antenna calculations will also treat it as an infinite ground plane. Sounds perfect.
However, after I run my simulation, I get directivity nearly twice lower than the gain, which of course makes no sense at all. I tried making the airbox larger, meshing it more fine, nothing changed.

Then I made an experiment and removed the "infinite ground plane" setting, so that it's just a perfect E square. After that I got results that make sense, with directivity higher than gain.

The interesting thing is - I compared the two results and found that the gain is mostly unchanged, and the directivity is twice smaller with the infinite setting on than without it.
So, what does it mean? Is HFSS computing something wrong, and in that case, which one - directivity or gain? Or is the problem with my simulation definition? Has anyone encountered something similar?

* For work reasons I can't post any pictures, so please bear with my descriptions.

Thanks ahead for any assistance.