I think my poor SCSI results are due to the drives I used, rather than the SCSI interface itself. The only SCSI drives I have handy are ZuluSCSI (limited to 8MB/sec max) and an old 1.2GB Quantum Fireball drive (even slower).
I tested a 40GB IDE hard drive in place of my original 60GB IDE, and surprisingly it performed much better despite being a nearly identical drive. They're both 7200 rpm Seagate Barracudas, the original 60GB is ST360021A and the "new" 40GB is ST340014A. For large transfer sizes, the 40GB performance was almost exactly 2x the 60GB drive across all test types. For smaller reads they were comparable, and for smaller writes the 60GB drive was faster. It's almost like the 40GB drive is two platters in parallel, and the 60GB drive is three platters in series, but the 60GB has more cache? Or maybe the 60GB drive simply has hardware problems, as I've suspected from the beginning. I'll post the full numbers later once I have all the drives tested.
I'm seeing many different drives max out around 55 MB/sec, so that number may be the practical maximum for IDE on this computer.
Whoever said that the second IDE channel on the Quicksilver was slower than the first channel was right! The 40GB Barracuda is like 2x to 4x faster on the main IDE channel (with the primary hard disk), compared to when it's on the second channel (with the optical drive). Bernie Sanders huuuuge difference. Don't put a drive on the second channel if you care about I/O performance.