Tradesman1 :
The freq of the DRAM - say 800 sticks running at DDR (DOUBLE data rate -= effective 1600 runs up to a theoretical max of 12,800 M/Ts a sec, and does this regardless of if it's single 64 bit, dual 128 bit, tri 192 bit, or quad 256 bit channel mode.
Looks like you have the wrong definition of what a MT/s is.
MT/s stands for Mega Transitions per second, which is equivalent to megabits per pin. 1600MT/s is 1600MT/s regardless of how wide the bus is, regardless of memory technology and regardless of what the base clock might be. Once you multiply it by the bus width, you get the effective bandwidth in Mbps which you then divide by eight to get MB/s. That 12800MB/s (or 12.8GB/s) figure is PER 64BITS CHANNEL at 1600MT/s.
bandwidth = effective bus width * bit rate
So, doubling the effective bus width (such as by upgrading from having a single DIMM in the system to having one DIMM per channel on a dual-channel board, which practically all mainstream boards have been for over a decade) does double available bandwidth. With a quad-channel architecture like Intel's Extreme chips, available memory bandwidth can be doubled again by populating all four channels since it doubles the effective memory interface width from 128bits for LGA115x to 256bits on LGA2011-*.
Multiplying bandwidth for a given memory technology by raising the effective memory bus width (by adding channels and making individual channels wider) is the whole point behind the progression from single-channel 32bits GPUs to today's multi-channel 512bits-wide high-end GPU memory interfaces.