There are generally two numbers given for memory, speed and latency. Speed is given as a frequency, generally in megahertz (MHz). This is the number of "clock cycles" per second the memory can handle. The higher the frequency, the more data can be pumped in and out of the RAM in an optimal situation (i.e. all data is lined up and ready to go, or similar). Due to fancy technological tricks, the various DDR memory standards (the most recent version is DDR3) can actually transfer data twice per clock cycle, so the actual "core" frequency will be half that listed ("1066 MHz" DDR3 will have a 533 MHz core clock, "1600 MHz" DDR3 will be 800 MHz, etc). This is important for the second number, latency. Latency is typically given as how many memory clock cycles of the memory it takes to access a random bit of memory within the module. This is important for DF, since the lower it is, the less time it will take to get the information about a random piece of rock or whatever. To get the total time latency, you divide the number of clock cycles by the clock frequency. As an example, DDR3-1600 RAM with a latency of 8 cycles will take 8/(800 x 10^6) = 10 nanoseconds to find a random requested segment of data. DDR3-1066 RAM with a latency of 6 cycles will take 11.25 nanoseconds. Note that the second example actually has slightly higher latency despite the lower clock-cycle count, since its clock cycles are slower.
High speed, low latency memory generally either uses a slightly higher voltage than normal (which may or may not wear our the memory faster/be slightly unstable, depending on the quality of the chips), is somewhat more expensive than normal memory (since they have to screen the memory chips to find the ones that run fast and reliably at normal voltages), or both. At the moment, I believe even good quality memory is still quite cheap, certainly much less than, say, a high-end graphics card.