The argument goes like this. There is very little real world benefit to raising the RAM speed on the core 2 architecture, assuming you don't have an Allendale with the small cache. So why bother driving the RAM hard and running the northbridge at other than a 1:1 multiplier, thereby increasing the heat of both needlessly. It's better to lower the multiplier of the CPU and attempt to achieve a 1:1 with the RAM at it's rated speed or to underclock the RAM to 1:1 and tighten the timings.
The 1:1 ratio really becomes an issue if the OC is unstable, because 1:1 more stable. If, let's say, you want to run a Q6600 at 3.0G and you have DDR2 800 then your only choices are 1333 (333)/800 or 1333 (333)/667 with tighter timings. There is very little difference between these to options in real world apps, and the RAM and northbridge don't really get too hot. So either way is roughly the same.
If you can it is better, obviously, to lower the CPU multiplier to 8 and raise the FSB to 1600 (400) and run the CPU at 3.2G with the RAM at 800 1:1.
If you have DDR2 1066 and a CPU that is running at 1333, then it makes sense to lower the RAM to 800 and tighten the timings and thereby lower the temps even though it's not 1:1 because the additional real world performance is negligible.
Additionally you could try 667 with even tighter timings at 1:1, but IMO the heat is no longer an issue at that point.
Many do not agree with me and really like the 1:1 ratio and tighter timings over all else.