Flandry: Is the CF is hooked to the same IDE channel as the system's optical drive? If this is the case that's probably why it's only running in mode 2, with IDE you will be limited by the slowest device on the channel, which is the optical drive.
The quick-n-dirty test setup is all kinds of wrong. It's on the middle connector of the IDE cable to the optical drive. Turning the drive off (unplugging the power) doesn't affect the DMA mode the card operates at, but
1) it's probably electrically detrimental even with the drive off (antenna effect) and
2) i'm not sure that the cable is an 80-conductor cable. I can't imagine why i would have used an antique 40 conductor, but i did put the system together in '02 and might have wanted to save the 80 conductor cable for a drive that would actually use the bandwidth.
I did do some reading on the ATA specification yesterday and devices are actually able to operate at different DMA levels on the same cable (i.e. that's an urban legend). The only time it would slow things down is when both devices are trying to use the cable. When i have the optical drive plugged in, it's at UDMA 2 while the card is at UDMA 3. The benchmark results of the card are identical in either case.
Copper: I remember reading somewhere that Intel chipset southbridges had a known problem with extremely low latency storage (e.g. CF/SSD). I can't remember where, but perhaps that's what's rearing it's head.
It just occurred to me that i could enable the built-in RAID and its two ATA channels and test performance on a different controller chip in an otherwise identical setup. Cool...
I'll report back.