Boobers: the amount of L2 cache has nothing to do with the amount of system RAM that is cacheable. That depends entirely on the cache controller. Modern processors (PIII, Athlon, P4, etc), which have onboard L2 cache controllers, have a 4 gig cacheable area I believe. The L2 cache controller on the AMD K6-3, could address "only" 512 MB. The cacheable area on the old Intel socket 7 boards (i430 FX, HX, TX) was only 64 MB! It is true that upgrading memory beyond the cacheable area can slow down the system because heavily used sections of code and data could end up in a non-cached page in memory. However, since the cacheable area of the newer processors is equal to the amount of memory addressable by the 32-bit x86 architecture itself, that is no longer a factor.
Plus, it DOES matter how many processes you run concurrently. If you run many apps simultaneously, adding more RAM will enable your system to avoid swapping programs in and out of memory during context switches. While this situation is mitigated by the fact that oft-used pages are not swapped to disk, some applications jump around alot in memory. If many of these applications are running, a phenomenon known as "thrashing" can occur where one app is swapped in, then another one, then the first one, back and forth, ad infinitum.