If you have to simulate something this large but need detailed modeling for the interaction, I would recommend you start with a time domain code; especially if you can get away with stimulating one transmit antenna and record the signals coupled to each receiving antenna. Then you only need one port simulation per transmit antenna.
Time domain analysis (FDTD, TLM, FIT, etc) will give you broadband response and memory requirements increase linearly with problem size; that is a pretty good advantage over finite-element frequency domain approaches.
Many academic codes exist out there for this kind of analysis. Some commercial codes would include Flomerics Micro-Stripes, Remcom XFDTD, CST Microwave Studio (actually time domain is just one of their solvers), and Zeland Fidelity; there are probably others as well.
-Max