Okay here we go and this is going to be a mouthful....
There are two sides of this fence, there are the pure CAS timing folks who equate lower cas times to increased memory performance, and those who believe CAS times are secondary to memory bandwidth by increasing the ram speed. To be honest in certain applications both are correct, so let me explain.
The first side of the coin is the fine tuning, or the "superbike" argument. By finely tuning your ram down to the lowest cas times possible(2 usually) you can ensure that the stobe fires and refreshes as quickly as possible on the ram banks, this will make absolutely certain that there is the least amount of slack during operations. In other words, the ram will function more efficently, much like what a supercharger does to a car engine. This will not increase memory bandwidth much, but it will take what you have and make it perform a hell of alot better then it would otherwise. This gives the illusion of more bandwidth, because more effecient ram, can handle more operations. The reason why low latency ram costs more, is because the faster cas times tax the substrate a lot, and the reason most ram can't handle this is because there is 0 margin for error. When ram is zipping along at cas 2 it's alot like racing down the road at 150mph, you can't make a single mistake. The reason why low latency ram doesn't traditionally overclock well is because of heat and signal noise. Super fast cas runs hot, and makes a crapload of signal noise, and even a single error along the strobe will cause a dump in the entire cycle and you will lose everyhing currently riding over that ram bank, ie it crashes. So keeping memory in parameters becomes essential. This style of fine tuning benefits quick small changes, like those commonly found in video games or the like. This is akin to a nimble machine, that stops on a dime and start up again just as fast.
The other side of the coin is pure bandwidth, or the "harley davidson" argument. By giving the CAS times some slack you are not asking as much of the ram in reguards to heat and signal noise, so you can punish it alot more before it cries uncle. By relaxing the CAS times, you can ramp up the memory clock signifigantly, because the extra heat and signal noise created by such a massive overclock, is tolerated by the ram simply because it doesn't have to pass the data around internally as fast. Much like a harley, they pour on the horsepower with minor performance upgrades and hope the extra horsepower makes up for the ineficiency of the engine itself. The same theory applies here. If you jack up the ram clock you increase the memory bandwidth available to the process with the hope that the processor will be swimming in memory and not need to pre/re-fetch as often. This stile of fine tuning is more suited to extended taks, such as editing, or content creation. Think of a steam shovel moving massive amounts of earth per movemnt, slow on the uptake and reload, but it can move a cubic assload of material in one motion.
I myself have found that to squeeze the most performance out of your ram, you need to find a happy medium between these two styles of memory tuning. Lower your CAS as far as it can go, and then try and overclock it, keep going till it becomes unstable. Once you hit that point back off and repeat the process until you hit the stability you are searching for. I've discovered on a good stick of Corsair or Mushkin I can keep cas2 and hit 240-250mhz on the ram, which is pretty god damned good. With such a high overclock running at cas2 my memory timings are off the charts. Also remember with lowered CAS times your can overclock your processor faster as well and keep the memory in line with that, so that is also something to take into account. Hope this helps, if you have any questions launch away.
