Okay, u asked for it
1. link concerning Ram timings =
http://www.tbreak.com/reviews/...pu&id=333&pagenumber=1
2. Now for that dual channel thing:
the problem with dual channel on Athlon XP was that the CPU bandwith was 2x200MHz FSB (if it was clocked to 200Mhz, like a 2500+@3200+niveau), and the Memory bandwith using 400MHz DDR Ram was 2x200MHz Ram clock in single channel aswell. So it was almost perfect already and the new feature of 4x200MHz memory bandwith was more or less pointless, except for some minor effects, eg that the latencies (which have no big effect on performce as we already know) were a lil lower. And that in the case the full memory bandwith was already used by the CPU there was no extra room for the needs of agp and southbridge bandwith. with dual channel this would work better, so that southbridge and agp would have enuogh bandwith even when CPU was reading/writing memory with full bandwith. These were minor effects but at least something.
With Intel its in theory a lil bit different. Lets do some math in detail:
This example is for a Pentium4 Northwood with 533MHz Fsb, which is able to read/write eg has a CPU bandwith to the Core of:
133 MHz FSB * 64 bit (single channel) * 4 (quadpumped) Datapackets per Clock = 34048 MBit/s bzw. = 4200 MB/s
With dual channel DDR 266MHz Ram the memory bandwith to the Core would be:
133 MHz MemoryClock * 128 bit (Dual Channel) * 2 (DDR) Datapackets per Clock = 34048 Mbit/s = 4200 MB/s
So as u can see the slow PC2100 seemed to be a good choice for that CPU, but it turned out that due to the architecture of DDR Ram the maximum bandwith of the Ram could not be reached and that faster Ram with a higher bandwith would lead to optimal performance.
Finally Intel increased the FSB to 800MHz that offers this CPU bandwith:
200 MHz FSB * 64 bit * 4 (quadpumped) Datapackets per Clock = 51200 MBit/s = 6400 MB/s
Now, there is no other way than to use 400MHz Ram:
200 MHz Memory Clock * 128 bit (Dual Channel) * 2 (DDR) Datapackets per Clock = 51200 MBit/s = 6400 MB/s
Still theres the problem, that faster memory would give slightly better memory benchmark results cuz the Rams maximum bandwith cant be reached, eg its only an perfect Team in theory.
By now u will be sayin: Told ya, this guy is nuts. First he says Dual Channel (DC) is crap and now he proves u need it and u need even higher clocked Ram for best performance.
As these math above in theory is right, it will differ in practice and wont give u much performane due to one thing and i hope u understand the subtext

:
"3DMARK is cool but i cant play it with my friends!"
When u take a look at the benchmarks in this german review here (sry, didnt find a english one):
http://www.hardtecs4u.com/revi...ntel_i875p/index16.php
u will see that there is only one benchmark that benefits much and this is Sandra Memory benchmark. And this benchmark is a synthetic, a theoretical benchmark meassuring what is the possible bandwith and not what is the real bandwith used by an application.
The term dual channel makes people think their memory is workink twice as fast as in single channel mode. but the results in the benchmarks show that its far away from doubling the performance in real life siutations. Dual Channel wont help u increasing ur fps in games, wont make ur video encoding, zipping files or compiling work significantly faster, i mean not that much that u will even recognize it unless using a benchmark.
I dont wanna advise not to buy Dual Channel, it surely gives a lil extra performance and has become a standard in AMD and Intel CHipsets, but the performance gain is rather small. But its better to buy dual channel and think: hey, i doubled performance!, then to waste cash for low latency Ram and think: i got the lowest latency, my comp must be the fastest in town.
Hope this helps ;-) Maybe not, my english sucks, this text must sound like written by a 5 year old, lol, anyways, greetz.
e: here a benchmark showing the effects of dual channel regarding Athlon XP Systems.
http://www.tweakpc.de/hardware...rce2_benchmarks2_2.htm