toolmaker_03 :
So is it that the memory really is not stable at that speed, or is it that the hardware really is not there yet for that speed?
from what it sounds like the hardware is not there yet right?
At this point of time, all the good points made here thus far are correct. There is no need for the additional bandwidth, and the memory controller in existing CPUs can only support such a high frequency.
The memory controller in the AMD Phenom II, for example, needs to run optimally at a ratio of 3:1 to the base clock of the memory. So a 2000mhz memory controller (by default) can support comfortably and stably only 667mhz (rounded up) memory, which when DDR'd, goes to 1333. To reasonably run 2800, you need 1400x3 = 4200 mhz memory controller frequency, which is impossible currently on that CPU. So you have to use memory clock dividers. If you use memory clock dividers, you are essentially accessing the memory (which runs really fast) at only once every few available cycles, and thus the total bandwidth of the memory isn't really utilized by the CPU.
Another BIG reason why faster memory isn't adopted is due to the DDR3 architecture. This architecture of memory isn't designed for >2ghz operation, and thus silicon yields are low for higher binned DDR3's. This makes the availability of fast DDR3's an issue, and no manufacturer of CPUs would bet their flagship CPU on the ill-supply of top, top binned DDR3's because they are very rare and command a premium. For the money of 2800mhz certified sticks, you can probably get twice as many sticks of 1600mhz DDR3, thus why the LGA2011 CPUs support Quad Channel instead of Dual Channel. In the graphics world, this issue is solved with the introduction of GDDR5 (quad data rate memory). Something like this is coming circa Haswell or beyond with the next iteration of DDR-something or other.