How about two radios in a box, one for 2.4G and one for 5Ghz, both mimo (prepare for the future, make it 3x3) and each chain should have at least one horizontal and one vertical antenna.
Write some software that swaps between the antennas (per radio) and test the radio link to the client device which antenna chain actually gives the best results (ack)
This is why Ruckuswirelss is performing so much better in indoor environments with their 9, 12 or even 24 antenna's (per radio!). And all just with normal 802.11a/b/g/n chipsets. No proprietary tdma or nothing!
Now Ruckus is very costly and patent protected. But I'll bet MT can make something that is going an end in their direction for much less money without getting the lawyers on their neck.
Any that want to know how real Wifi should work should take a look at www.ruckuswireless.com.