The Guide that CanadaRAM refers to is based on information by CanadaRAM, so its kind of circular. However, Ive seen a variety of posts that say the same thing, and I think what he says is correct.
Everyone seems to agree on the following: 2x1GB works in all Intel Macs, and provides dual-channel access which is a little bit faster than a single 2GB module. In Core2Duo MacBooks, 1GB+2GB works and provides extra memory but not dual channel, which is a net gain if you would get any page-outs with less RAM. In Santa Rosa MBPs, 2x2GB provides dual-channel access to 4GB. In C2D MBs, 2x2GB works, and provides dual-channel access to 3.3 GB, with the rest being unaddressable because the addresses are already being used by the system.
Some questions remain:
(1) There was a report here recently that someone put in 2x2GB in a C2D MB, that the system saw all 4 GB, and that it led to crashes. (I tried searching for the thread, couldnt find it.) I think that involved using a low-level tool, and that the tool was being used to address memory locations outside the OS controls, which led to the crashes. So, if you put in 2x2GB, and dont play with low-level stuff, will the system stay happy? Will you get memory crashes and conflicts if you put in 2x2GB?
(2) Does the 3.3 GB include the graphics memory, or is the graphics memory part of the 700 MB that the system is using? It sounds like the OS really cant see those memory locations at all, so the graphics memory should come out of the 3.3, I think.
(3) In a Santa Rosa MBP, is 4GB a hardware limitation, or is it just that we dont have 4GB modules yet? That is, in the future, will Santa Rosa MBPs be able to use 2x4GB?