Yup, soundstaging evaluation is difficult indeed. Used to work with HD590s, but now I'm spoiled by decent-sounding supraaurals.
A look on the card: Opamps, where present, are JRC 4558. The power amp is a LM4880M. 49.152 MHz main xtal. DAC PCM1725U, as mentioned. Didn't see any dedicated ADC, might be done by STAC9704. Coupling caps are 470µ 105° (but rather small) on front out and 10µ25 on rear out. Power buffering mainly done with 10µ25. 2 sets of jumpers for line <-> headphone out switching - one to disable 4558, one to enable LM4880? Overall, a state of the art prosumer sound card of 1999. Hadn't the Win2k driver issue been, it could have blown away the SB Live!.
Edit: I thought the occasional crackling was related to PCI latency, but apparently it may not be. I'm at 512 samples and 4 kernel buffers now, let's hope that this at least works as well as WaveOut which only very occasionally produces crackling. I've increased the latency timer for the DMX to 64 clocks (via the Audiotrak PCI Latency Tool) in addition, just to be sure.
The writeup below may be a bit moot now but I didn't want to delete it either:
Gotta love crummy old PCI implementations. (PCI video on the 440FX without write combining enabled is no fun, I tell ya - and while the default Win2k driver for the Millennium II does that, the newer one from Matrox does not or not correctly. Believe it or not, this was the first PPro chipset where PCI actually worked halfway well, the preceding 450KX/GX needed a few revisions before PCI acceleration features were sufficiently bug free to be turned on by default - before, throughput was limping along at 8 MB/s, and given that PPro machines were high-end stuff at the time - we're talking 1996 here - this pretty much sucked.) Of course it's not unlikely that the ESS chip isn't entirely innocent - someone please name a first-generation PCI audio accelerator that worked really well all around.