>Is this actually true?  I thought modern IDE interfaces were just as
>capable of "start it and forget it" operation as any SCSI interface.

Does UDMA realLy allow disconnect/reconnect? On buses with drives
of mixed capabilty (as SCSI does?) How about per-target sync speed
(for IDE "slaves", can slave can do faster than  master)?

that aside, a Promise UDMA?66 controller is a whople lot cheaper
than a SCSI-160 controller, but with a string of IBM LZW glass
drives tha latter can (I beleive) sustain more. 
Again, corrections welcome.

10,000 RPM high-capacity  IDE drives are, currently, Unobtanium.
But so are are 1in 40gig SCSI drives.

>(And don't forget that SCSI interfaces demanding PIO exist, though
>quite possibly not for i386 machines.)

Any hardware botch that can be perpetuated, has been on i386.