<< but the reliablitity problems are probably because they either made their drives to fast or they spent too much time trying to make them fast and didn't spend enough time on reliability >>
Well... no.
The issue with the 75GXPs is platter expansion. In all HDDs, data is written to a platter by a write head changing the magnetic orientation of very small areas on the platter. Under normal operation, this works perfectly and allows us to store data to and read data from our HDDs.
Unfortunately, sometimes things go awry. In the IBM 75GXPs, the substrate of the platters, the actual non-magnetic material that makes the platters themselves (platters are made of aluminum or glass, for instance, then coated with a magnetic material) expands when a level of heat is reached. Now, we're not talking about expansion on a scale large enough to cause cracks in the magnetic film. Just a tiny amount...
But that small bit of expansion is enough to totally throw off the read heads, which expect data to be in certain places on certain platters. Said data is still there, but due to the expansion, it's been shifted out of its original physical position ever-so-slightly. The result being, the read heads fail to find the data they're expecting to find and reset their positions... over and over again.
*click click click*
Once the drive cools (read, has been off for a while), the platters contract to their normal size again, and the data can be retrieved. But each time the drive is used, it heats up and causes platter expansion again, making it unusable in a normal working environment, under normal working circumstances.
The problem is exacerbated by ferromagnetic-resistance heads, which make use of electron spin to read data. This works by having a platter material in which there's a global electron spin orientation. When data is written to the magnetic film that coats the platter, it causes the spin orientation in the precise area where data's written to change (either opposite that of the platter, or matching that of the platter) and create either electrical resistance (opposite spin) or permit current to pass through unhindered (matching spin). It's this measure of resistance that determines whether each bit is a 1 or 0. While this allows greater areal densities, it also leads to problems like what everyone's experienced with the 75GXP line.
IOW, the closer you pack the data on a platter, the greater the problem when the platter expands and the more sensitive the heads become to slighter and slighter expansions. Everything's still there, but it's just not where it's physically _supposed_ to be, which makes it irretrievable until the platter contracts to its former dimensions.
And that's why the 75GXP is a cold deal.
Any questions? Anybody? Anybody? Bueller? Bueller? ;-)