Ok, time for some lessons. I've posted this other forums before, so I'll quote myself:
Additionally, I think there's a few misconceptions in this thread. For one, "rounding" your own cable actually PROMOTES crosstalk for a very good reason. Normally you have a data line seperated, then a ground wire, then another data line, etc. When you round your own cable, you're folding all of these together so you get data lines right against one another without being seperated by those ground wires. This is how you get "crosstalk" on the same cable between data lines. Length of the cable has very little to do with crosstalk.
However, what length of the cable DOES impact is "impedance matching". Unlike SCSI, IDE cable connections are unterminated. "Termination" is what you have to do to stop the signal from reflecting off the end of the cable like a wave that hits the end of a bathtub. The longer your cable, the more end-to-end signal delay you get in terms of fractions of a clock pulse. At UDMA/100 speed, this reflected signal is more and more out of step with the real signal than at UDMA/66 or UDMA/33. Because most drives are now either UDMA/100 or UDMA/133, they are more susceptible to cable problems. That's the whole reason for the 80 wire cable standard anyway, to make the cables less error prone by adding all those interleaved earth (grounds) that are tied together at each end. Ribbon cables with interleaved earths have easily characterized impedance.
With interleaved earths, each signal wire interacts very little with the signal wires on either side. The electromagnetic and electrostatic fields fromt eh signal wires get eaten by the earths in between them. Remove the earths, and the cable impedance can vary tremendously (ie by a factor of two or three) depending on what wires are and aren't energized at that moment.
IDE device designers don't know how the cable's going to be routed, wheter it's going to be hanging in the middle fo the case away from the chassis metalwok, or tightly clamped to the side of the case, or some mixture of both. This determines the cable's capacitance to earth, since the PC case is earthed, and that also influences the impedance.
What's all this mean? If the impedance is lower than the "driver" circuitry that's trying to send a signal expects, then the driver will be unable to deliver enough current at full signal voltage, the signal strength will thus fall, and you're on the way to data loss. Data-signal drivers have much less trouble feeding high impedance loads, but whn the impednace of a 40 wire cable rises for whatever reason, the impedance of the drive on the end of it doesn't. That creates an impedance mismatch at the end, which causes signal reflection problems. Ideally, you don't want the impedance of the cable to vary at all, and the interleaved earths among the signal wires does a pretty good job of squishing the impedance fluctuations.
So, what happens when you split an 80-wire interleaved earth ribbon cable into individual wires and bundle them up? The mixture of data and earth wires is not going to give you a tightly characterised cable, and you now have the problems above I was referring to. IDE spec is drawn out at 10" minimum, 18" max for a reason.
Now, for those of you saying, "You're full of it! I've run longer cables and round ones for YEARS with no problems", I'll just say that can be explained. Good enough components at each end of a cable can deal with more signal corruption than the IDE specification demands. Modern ATA hardware is pretty good at dealing with lousy cables. Current IDE hardware can shout through the noise fairly well. Secondly, IDE covers up data loss problems in that drives mostly have CRC error checking built in now. When data's munged in transit down the cable, the error is detected and the data is resent.
Well, USUALLY it is. I know myself I've got more problems ripping music with my Toshiba DVD-Rom than I do with my Sony Dru-500A DVD-writer. Using Fuerio, I've seen CRC errors on ripped music that if I re-rip with the other drive I don't have. These are ones that CRC CATCHES......
CRC will always catch a one-bit error in a given data block. But if MORE than one bit is wrong, the CRC data may end up the same as it'd be if everything were fine. That's not likely to happen to any given block of data, but in a system with serious data corruption problems it certainly can happen rather often. Hard drives move a lot of blocks of data in a day. This can give you corrupt data on the drive or to your memory, and/or a crash, depending on which way the data block was going and why. Even if you trust software CRC checking, you don't want a lot of errors as it slows down your drives. Remember, CRC only provides error detection, not error correction. The data must be sent again.
This is also why as a rule you don't want to overclock your PCI bus because you also will be overclocking your IDE bus as this can increase the error rate.
If you have a major IDE data loss problem, things are obviously broken. Drives won't be recognized consistently (or at all) on boot, every file operation of a significant size will cause errors, swap file activity will hang the computer. Not the kind of problems you're likely to get from say a standard over-length cable, though. Mostly, you would just see slow drive performance if the CRC is doing it's job.
Since your swap file is the most heavily accessed part of your drive (depending on RAM amount and application), IDE data integrity problems can cause the same symptoms as faulty RAM. Also, think about this...what's the second most accessed single chunk of disc space on Windows? If you said "registry", you win a cookie. And you REALLY don't want data corruption there.
I took all this into account when I put together my system. What did I use? Rounded 20" and 12" cables. At this point, you're probably saying, what the hell?
Well, 20" isn't TOO far out of spec from the IDE length limits, and I wanted rounded cables for airflow reasons, same as most people. However, I also wanted good data. What with ever increasing bus speeds, soundcards, video card frequencies, a TV tuner, CPU, etc., there's a lot of signals bouncing around in your case, and those long IDE lines start looking a lot like antennaes (which if you think about it, they are). So, I ended up getting IOSS RD3XP Gladiator cables for my system for a couple of the above reasons. Most notably, these cables are fully shielded with a ground wire. When I say fully, I mean the entire outside is foil wrapped underneath the mesh to help protect from stray signals in your box. Secondly, even though these cables are rounded, they're not constructed like your garden variety rounded cable. The cable is layed out in 8 layers of 10 wires, alternating ground and signal on each layer, and also folded in a zizag pile so each signal wire is surrounded by 4 ground wires. Not only that, between each of the eight layers of wires is another aluminum foil shield. That's also an added reason I thought the 2" over IDE spec wasn't as big a deal.
Anyway, that's some of the reasons I DON'T recommending folding your own (or buying most rounded cables for that matter, with one obvious exception since it's the only one I know of that shields their cable and preserves the layout design idea of an 80 wire cable).
And yes, as a general rule, using master / slave settings on the drive rather than cable select is a good idea. Most will also tell you the burner should be master. I personally wanted the drive I reach the most (at the top) to be the DVD-ROM so I only use the burner when I want something burned / ripped. That necessitated seting the burner as slave, and I haven't had any problems, be it flashing firmware or burning a cd/dvd.