Latency is only an issue when both the optical and hard drive are used at the same time. As you say, commands in the same IDE bus may not be overlapped. Provided the optical drive is not used, there should be no speed/latency penalty associated with just having it attached to the cable.
We know from golgos post number 26, that both drives use the SCSI/libata drivers as /dev/hda and /dev/hdb are both missing. This implies that DMA is always on. There is no SCSI/IDE translation in this process, its not like the old ide-scsi driver that used to be needed for burning under 2.2 and 2.4 kernels.
I have heard anacdotal evidence that the old IDE driver branch is better that libata for ATA drives but testing that is a kernel rebuild. Its not likely to fix this issue either.
Some further tests.
Is your system slow when you do not use the CDROM drive ?
If so, disconnect it from the IDE data cable and test again.
If your system is still slow, reboot it and post the output of