Is it faster to have four times 2 GB or two times 4 GB of RAM with a dual-channel mainboard?

"Four sticks would cause more strain on the memory controller and motherboard chipset. It would take slightly longer for the CPU to add and retrieve data from four sticks as opposed to two.

For this reason, 2 x 4 GB would be faster then 4 x 2 GB.

EDIT - there is a much better technical explanation on The Hyphenated Site to enforce my answer, although it does mention lower sized sticks:

It's better to use two 2 GB modules --
not for any appreciable speed
difference (although there may be a
small advantage -- more in a bit) --
but for a more reliable memory
subsystem.

Most desktop systems use unbuffered
RAM modules -- this results in very
large loads on the address and data
buses when you have more than two
modules installed, and can
significantly degrade the signalling
on these buses. The memory subsystem
""sees' one load per memory chip -- so
with two modules installed, that's up to
32 loads (with double-sided modules)
... and with four modules installed
that's as many as 64 electrical loads
on the bus. Some systems
automatically adjust for this higher
load by either increasing the voltage
a small amount; reducing the clock
frequency of the memory (thus slightly
slowing it down); or by adding a cycle
to the SPD's latency setting (again,
slightly slowing it down). These
adjustments help keep the memory
subsystem reliable -- but mean that 4
x 1 GB modules would be slightly slower
than 2 x 2 GB modules on these systems.
But regardless of whether there's any
timing differences, the memory will
definitely be more reliable with only
two modules.

Credit to garycase at The Hyphenated Site for this answer."

"Not my answer but the one that I followed:

This question has come up before in several other forums, but there has never been a good clear answer. Overclocking, upgradability, and temperature aside, I would suspect that 4x2 would be slightly faster than 2x4 because there would be more interleaving. Interleaving means that the data is spread out across more memory chips. While some chips are waiting for their CAS timing cycle to complete, data access can occur from the other chips.

The thing is, it really depends on the northbridge (NB) implementation. The NB must be able to take advantage of interleaving across four DIMMs, and I don't know which NBs do and which don't. I suspect the most of the Xeon NBs can interleave across four DIMMs, in fact most of those NBs require DIMMs to be installed in groups of four. I'm not sure about the consumer NB lines such as P35, 965, and nForce 680i, the answer could be different for each chipset...

I know this does not directly answer your question. I am quite certain that 4x2 will never be slower than 2x4, it will either be the same or faster if the NB supports four-way interleave."