FSBs and memory speeds seem to be one of the more often misunderstood aspects of computer hardware, and it doesn't seem that any of the answers submitted so far satisfactorily address the OP's question, so I will throw in my proverbial 2p.
FSB : DRAM Bandwidth Matching
All current Intel systems feature a 'Quad Pumped ' FSB, that actually runs at a quarter of its rated speed. It is a 64-bit bus that connects the processor to the Northbridge. The memory bus connects the Northbridge to the RAM, and runs at an actual clock speed of half the rated memory clock. In dual channel mode, we have two 64-bit memory modules (effectively 128 bits, or 16 bytes) running in parallel.
An '800MHz' FSB can provide 800 * 64bits (8 bytes) = 6400MB/s of bandwidth to/from the Northbridge. So, to provide the FSB with as much bandwidth as it can handle, we need two modules of RAM running at an effective speed of: 6400MB / 16 bytes = 400MHz (DDR400).
Therefore, the general rule when matching FSB speeds to RAM speeds is simply:
RAM speed = Effective FSB / 2
This is not to say that installing memory capable of higher bandwidth than the FSB is not a good idea. In fact, memory running at higher speeds tend to have lower latencies, so you could see some benefit in running DDR2-800 with an FSB of 1333MHz, as opposed to DDR2-667 (assuming the same CL).
For those interested, 'Quad Pumping' is achieved through the use of:
1. Double Data Rate (DDR) transmission, whereby data is transmitted at 45 degrees and 135 degrees of the sine wave. This is the same method used by DDR RAM.
2. A second signal, transmitted 90 degrees out-of-phase with the first.
Taken together, we have two signals transmitting data twice per clock, giving an effective FSB speed of 4X the actual.