Our lab is using USRP N310's with an external frequency synth ($500), a TI dev board splitting chip ($200?) and an external 10 MHz selenium clock ($2k). The set up is very very good but extremely expensive as well.
After paying for RF cables and a host server, we are roughly getting 1 channel per $4k.
For MIMO and beam-forming, it might be cost effective for you to buy more N210's in order to use the hardware you already have.
A wireless research lab on campus uses lots of B-series USRP's because they are a (relatively) cheep way to set up lots of AP's and clients. That lab has a good 20-30 B-series devices and used them for a wide variety of experiments. Not sure if they are suitable for your desired experiments.
Alternatively, this non-Ettus device may be worth looking into:
I would like to warn you that RF cables (SMA. BNC, N-connector) are surprisingly expensive (and effected by the American tariffs). You should budget accordingly.