I have a similer system to yourself. I have an AthlonXP Barton overclocked to 3200xp rating on an Abit NF7-s (nforce2) motherboard. The Nec is on IDE channel 2 as slave (to the Liteon).
My cd-rom CPU loadings are
1x - 0%
2x - 1%
4x - 2%
8x - 5%
----which are the same as the cdfeak tests I think.
I tried testing a Cd-R and it crashed. Then I realised that the only cd-R i'd made was of a protected game so some of the sectors are probaby unreadable!
I tried using the NEC by itself on IDE channel 1 and it didn't change the results.
As far as I can see the Nec is producing the results I would expect from a DMA enabled device except for the Burst mode figures. It's as if the Nec's buffer memory is running slower than the interface spec. But this can't be the case as 1mb/sec is slower than both the maximum read and maximum write disk transfer speeds. I'm beginning to believe that it may come down to how the burst mode test is performed.