Fujitsu's recent problems were not from an overheating chip, but rather, one packaged in a bad epoxy (or some such thing) which had the unfortunate effect of reacting with the silicon. Hearsay among friends battling the Deathstar 75GXP issue pointed to the 'pushing' of the electronics and firmware basis standard among the model line; supposedly the on-disk controller couldn't keep up with the data rate from the high-density platters in all cases.
I made the mistake of buying a WD800AB a while back, which appears to have some issues relating to power consumption, the automatic parking feature meant to kick in on voltage (and thus RPM) sag, or just life in general. I never did get a chance to test it in a machine with a beefier supply.
Point is, it's tempting to blame heat, but we're equally up against complexity in general. It's also amazing how IDE *hasn't* evolved; sure, some drives support TCQ and automatic remapping, but given the utilities provided with the WD, I'm fairly sure it lacked even the latter. (I'll confess to never having found a proper SMART util to use with it, as well.) Think about the tolerances necessary for today's data densities, and I'm fairly sure the mechanicals are an issue as well. The prevalence of cheap RAID certainly can't be helping funding for reliability R&D.
Meanwhile, I can't seem to kill the noisy, hot, slow-for-7200RPM refurb Barracudas I picked up 5 years ago, despite the vibration of multiple road trips and their current swaddling in paper towels and packing tape.
---
And, uh, the article was written under the assumption that we all know HD latencies can be killer, and that 'flat' high-speed memories (MRAM?) could be the holy grail for a number of applications. If anything, the conclusion that it's not as bad an issue as some might think is surprising.
