I have been asked to see if there is a way to reduce the necessary number of antennas from the obvious 8, down to 1 or 2 while still keeping all phones fully utilised simultaneously all the time (e.g. all 4 phones are running 90% duty cycle or more, all the time).
The question whether you can remove the numbers of antennas is pretty much up to the device manufacturer – however, MIMO basically works because your antennas receive something different. Talking very abstractingly, a MIMO system uses a Matrix representation of your N transmit antennas – to – M receive antennas system, and finds suitable decompositions of that matrix so that the amount of data transportable can be maximized.
If all but one receive antenna are deaf, the whole MIMO collapses to MISO, but all algorithms that actually get deployed somewhere should still work – it's not unlikely that one antenna has an SNR that is so much better than all the others, that diversity gain is negligible.
The same goes for TX.
The only thing I can see for TX is that with an unconnected antenna port, you might damage the TX amplifier with the reflected power. But that's something that the manufacturer of your devices will be able to tell you. In any case, you should alternatively be able just to terminate these connectors with a matching terminator – again, the MIMO would collapse, but SISO is just a "special case" of MIMO, if you will so.
Combining multiple antennas into one antenna doesn't make sense – as said in the comments to your question, for MIMO to do anything, it's crucial that the receive signals are diverse, which means if you just used the same antenna signal for all 4 channels, you'd win nothing; the only thing you'd do is split your receive power by 4, and that will reduce SNR. So don't do that.
The main reason for thinking about this problem is that we don't want to route out 8 antennas when we can get away with 1 or 2.
The main reason I'm not driving a Benz is because I can get away with a bike. However, if I had a Benz, I might be faster, have less outage probability and be more efficient at work.
This whole idea sounds REALLY expensive to me, but lets ignore that minor issue for now, I'm mostly after a feasibility check.
Why would even your approach sound expensive? The power splitters/combiners you mention would be easy to realize in microstrip on a PCB (in fact, just as a couple of antennas.
Personally I'd rather just run out 8 antennas (nice and cheap and definitely works) but this system needs to be scalable up to perhaps 12 phones (24 antennas would look like Cthulu's grandma!).
A nice old lady. Has the best cookies in town, but once you're there for tea, you gotta listen to stories about her grandson until your ears grow back and you develop scales. Anyway:
Really, I'd go with your approach, here. Make use of let's say two patch antennas (cheap as hell, not very large for 900 MHz or 1800 MHz or so), spaced at least half a wavelength apart, and two inverted-F or something more "linearly" shaped so that you get waves from different directions.