How many SS' an AP can transmit (or receive for that matter) depends on the last parameter in the TxR:S notation. If the S parameter is 2 it can send 2.
However, if you have a 3x3:2 AP it can send those 2 streams over all 3 radios. There are various reasons why you would have more radios than spatial streams (polarization diversity, STBC, others) and vendors implement different features for that. Each radio has it's own antenna.
One thing to clearly remember and understand is that an AP is a HUB NOT a SWITCH(say it out loud a couple of times). It will send data packets to one STA (client or another AP) at a time and it will expect an acknowledgement for each one.
802.11ac will change this a bit in the future. 11ac started with the MU-MIMO idea, that, as you sad, will send packets to more than one STA simultaneously, however it will require more than 4 radios to be of any use there.
In the case of an iPhone, which is a 1x1:1 device, a 2x2:2 AP can send one stream only to it, however it can leverage all of it's radios and send packets via all of it's antennas. Why?... well it can use beamforming (TxBF not RW adaptive antennas), STBC,...
AND it can also receive on all radios. This is what RW does that puts them miles above the rest. It has dual polarized antennas (one antenna is perpendicular to another), so it can receive signals that come from any direction and combine it positively. This is called MRC or as thy call it PD-MRC which stands for polarization diversity. They claim it adds up to 4dB of gain, which is big and it helps with the speed of the link and especially with transmissions that might occur if the client is at an edge of a cell for example, where another AP might fail because it doesn't have those 4dB of help. It means the AP hears much much better. In fact RW APs hear like dogs, while others like my grandma.
If everybody hears each other better the speed simply goes up, bar some other limitations that are usually the fault of bad WLAN design.
