Ten years, man...
The hard drive was a Toshiba P300 3TB model HDWD130XZSTA. In the last year or two, since Toshiba's corporate troubles have gone on, they have started to price their hard drives more competitively. They seem like pretty nice drives in regards to performance and noise, but I discovered that this model at least seems to be susceptible to errors when subjected to nearby acoustic noise/vibration.
The server was a generic 1U with six 40mm fans in the center behind the hard drives. When the fans run at full speed they are really loud -- like seriously hurting your ears loud.
I first noticed something was wrong when some SMART errors developed on one of the drives, indicating a failure. I RMAed that drive and then got a new one, but the linux mdadm RAID rebuild was insanely slow. Further tests showed that the new drive was performing terribly and I went through the process of troubleshooting.
It took awhile before I finally figured out it was the fan noise causing it, but you could really see the transfer speed ramp up/down when I plugged the fans in/out. When the fans running at full speed, the drive read rate (hdparm -h) would drop down to 1MBps or even less, whereas they would go up to 175-190MBps normally. The read errors were SO bad that it must have triggered a SMART failure on the previous drive.
I've been a sysadmin for about 20 years now in datacenter environments and this was the first time I've seen an effect this substantial.
It's noteworthy that this same chassis has two Seagate drives in it which didn't suffer any negative performance issues from the noise.
I guess that's what I get for using desktop class drives in a server.
I don't know how fans actually implement variable speed. But I can think of several methods that would generate lots of electrical noise. Fan manufacturers have an incentive to build as cheaply as possible, so I doubt they take too many steps to mitigate crap they put out on the power lines and/or radiate from their speed controller.