The rMBP has 2 ports and a single TB controller (single bus). Each port has a capacity of 20Gb/s. However, the TB controller only takes a single DP1.2 input from the GPU that can then be output on either of the two ports. This allows you to drive a 4K display using one port, and a storage device on the other. Or you can run two lesser displays (one on each port), or using MST, drive two lesser displays off a single port (using daisy chaining).
MST was designed to drive dual non-4K displays off a single DP1.2 connection by daisy-chainging the displays together. It effectively mux's two display signals onto a single DP1.2 connection.
When early 4K panels came on the scene, there was no single piece of silicon that had enough bandwidth to drive that many pixels at 60Hz, so early 4k panels utilized MST - treating each half of the panel like a separate display. This was a great workaround, but it's plagued with issues. For example, every panel can implement MST slightly differently requiring drivers on the Mac to support the different implementations. Plus MST can have issues with blanking one half the display or the other or waking from sleep. Not surprisingly, it's a kludge and to date, Apple officially only supports 4 different MST displays.
SST has only recently become possible as display silicon capable of driving a full 4K panel at 60Hz has become available. Now a single connection can pass the full DP1.2 signal necessary to drive the whole display at once, eliminating a lot of issues with the promise (at least) of allowing for a standard display driver that can simply use any display's EDID info to drive it effectively.