i have managed to duplicate this problem verbatim:

http://mail-index.netbsd.org/port-cobalt/2004/09/16/0003.html

on an ultra5 with a promise pdc20269-based card, 80-pin cable, uata133 disk,
and netbsd 2.0-rc2. so i'm guessing that the card gets initialized at bios
time on an x86 box, which obviously isn't happening here or on a cobalt.
has anyone researched this yet? any ideas where to start? i'm not above
disassembling the prom...

it's an incredible testimony to the netbsd developers that this works at
all, but it doesn't solve the, uh, performance challenge of an ultra5's
on-board ide :)