Not sure if that can help, but we've got that board on ultimateretro with pictures https://www.ultimateretro.net/en/motherboards/1481
And we also have one bios. I don't remember how it was dumped however (may be worth to check that it's allright).

My mobo looks different to the pictures, it's a perfect match for the drawing on TULARC - that being said there's a couple of clones with that name (the only thing written on the PCB is "386WB" behind ISA slots). You'll be happy to know the BIOS image on Ultimate Retro has correct cheksum, looks properly dumped. It's a 05/05/91 with a dozen or so bytes different to mine so I'm not expecting anything. Will test it later though.

In the meantime the plot thickens:
* AMD D0-step CPUs do usually boot with Intel 387DX, and my own benchmarks seem to run fine, will re-test with Quake to be sure
* AMD D0-step did boot twice with Cyrix FasMath, didn't test more yet (only so many hours in a day)
* TI 486DLC seems to actually work properly with Cyrix FasMath, but I only tested for the obvious hangs, no proper benchmarks were run
* seems I didn't test the combination of TI 486DLC with Intel 387DX, need to do that later as well

I have a B1 step of Intel 386DX but it's a 16MHz rated CPU, it'll only work for some 10s or so on 33MHz clock (and only on first boot when cold), then system becomes too unstable to run anything, or even boot. But I did manage to run a simple freeze-or-not test on x87 detection code and it didn't. Curious, one would think if a timing bug exists in D0 stepping, it would also affect earlier models. Maybe it's D0-specific? I mean Intel went through D0 and D1 before D2 became the final stepping of 386, they surely had a reason to re-spin the D series twice.

Another update, hopefully this helps someone with similar issues.

While doing DLC/SXL chip tests I was using active cooling not to stress them, and then re-used the setup for some 386DX chips. That was a mistake, it made a difference in the results because it seems this is indeed a timing issue between CPU, NPU and the chipset and the temperature does affect things. I had to re-do some of the test to verify this, here's the results:

Am386DX/DXL-33 [0305] - no cooling (~45C):
* Intel A80387DX-33
some apps work but Landmark and Quake hang
* IIT 3C87-40 / IIT ITT 4C87DLC-40
BIOS will usually freeze at POST on math co-processor detection code
* Cyrix CX-83D87-40-GP FASMATH
Landmark FPU score is unstable (you can watch the bar jump in realtime), Quake freezes, my own test shows glitched values

Am386DX/DXL-33 [0305] - active cooling (heatsink+fan):
* Intel A80387DX-33
everything works OK
* IIT 3C87-40 / IIT ITT 4C87DLC-40
kinda works, but not stable
* Cyrix CX-83D87-40-GP FASMATH
Landmark works, Quake will exhibit geometry glitches

Intel A80386DX-16 [0308] (33MHz) - no cooling (~50C):
* Intel A80387DX-33
everything works OK
* IIT 3C87-40 / IIT ITT 4C87DLC-40
everything works OK
* Cyrix CX-83D87-40-GP FASMATH
Landmark and Quake hang right away but it has to be noted this is overclocked CPU, so this is not conclusive

Cyrix Cx486DLC-40GP [0421] (33MHz) - (L1 -m-) no cooling (~60C):
* Intel A80387DX-33
everything works OK
* IIT 3C87-40 / IIT ITT 4C87DLC-40
everything works OK
* Cyrix CX-83D87-40-GP FASMATH
everything works OK

TI 486 SXL-40 [0421] (33MHz) - (L1 -m-) with heatsink (~55C):
* Intel A80387DX-33
everything works OK
* IIT 3C87-40 / IIT ITT 4C87DLC-40
everything works OK
* Cyrix CX-83D87-40-GP FASMATH
everything works OK

TI TX486DLC/E-33GA [0421] - (L1 -m-) no cooling (~65C):
* Intel A80387DX-33
Landmark FPU score is unstable, Quake hangs, so do eventually my own benchmark programs
* IIT 3C87-40 / IIT ITT 4C87DLC-40
all code trying to use NPU will hang the CPU bus, might even freeze at POST
* Cyrix CX-83D87-40-GP FASMATH
Landmark FPU score is unstable and it hangs eventually, Quake hangs

TI TX486DLC/E-33GA [0421] - (L1 -m-) with heatsink (~55C):
* Intel A80387DX-33
Landmark looks OK-ish but Quake still hangs, my own benchmark programs do seem to work OK
* IIT 3C87-40 / IIT ITT 4C87DLC-40
all code trying to use NPU will hang the CPU bus, sometimes a program will work for a bit though
* Cyrix CX-83D87-40-GP FASMATH
Landmark FPU score is unstable, Quake hangs

I should have another (ULSI brand) NPU in a week or two, might test this one as well. In general I would not run SXL chip without any sort of heatsink so that one was tested with it, other chips don't get that hot on their own. I would also not mount any sort of cooling, even passive heatsink, on a 386DX. And keep in mind this was all tested in free air, not in a closed case.

Cyrix chip is problem-free, and I would still put a heatsink on it at the very least if it was to end up in a case. So is TI SXL with heatsink, although for 40MHz bus and closed case I would probably consider a fan as well. Usually I do run it with both even during tests just to avoid stressing and heat cycling everything (it also doesn't help that the usual mobo for testing 386 CPUs that I use had the on-board PQFP 386DX and chipset right next to each other and that tends to really build heat up in that one spot).

Intel D2 stepping seems problem free except with Cyrix FasMath but that can be due to the overclock. Curiously all the chips that worked perfectly are 40MHz rated, running in 33MHz system. I wonder if there's anything to it. Chips rated at 33MHz or less gave me trouble - but only if the co-processor is installed. I didn't notice any instability in Doom tests or otherwise, and I did try less aggresive cache and DRAM timings.