Your doom timedemo result is very interesting.
I also have a 386 with an OPTi495 chipset, however mine is the 495SX not the 495SLC. The only real difference is that the 495SX has buffered VLB slots...they are more compatible, but they run slower. My board does not have VLB slots though.
My system configuration is:

-Generic OPTi495SX motherboard, 256kb Writeback cache
-TI486SXL-40 (a clone of 486DLC with 8kb L1 cache), Fasmath 83D87-40
-32MB FPM DRAM
-CL-GD5434 ISA video (Diamond Speedstar64)
-AHA-1542CP SCSI
-Yamaha OPL3SAx sound
Doom timedemo 3 score: 5891 (8MHz ISA)
Doom timedemo 3 score: 5592 (10MHz ISA)
You'll notice my score is significantly slower than yours...but that doesn't quite tell the whole story when you factor in the VLB video.
I think another interesting system to compare to is one based around an IBM 486SLC2-66. This CPU only has 16-bit datapath and 24-bit memory addressing like a 286, but it has 16kb internal cache. My system uses the OPTi295SX *with* VLB slots. As far as I know it is basically the 16-bit version of the 495SX.
Configuration:


-Alaris Leopard Rev C motherboard, 128kb cache
-IBM 486SLC2 66MHz (basically a 386SX with a lot of cache), Fasmath CX83S87-40JP
-16MB FPM DRAM
-ET4000W32P on VLB, or CL-GD5434 on ISA (Trident 9440 in the Speedsys capture)
-AHA-1542CP SCSI
-Yamaha OPL3SAx Sound
Doom timedemo 3 score with ET4000W32P on VLB: 4730 realtics
Doom timedemo 3 score with CL-GD5434 on ISA: 5284 realtics
The timedemo score is about 4730 no matter which VLB card is used. It appears that any VLB card will offer at least a 500 point advantage over the fastest ISA card (The CL GD5434 is the fastest one I am aware of).
Your 66MHz Cyrix DRx2 with 1kb cache is surprisingly slower than my 66MHz IBM. Now I could understand this being the case if my CPU was a DLC2, but mine is the SLC2 with the crippled 16-bit databus. It does have quite a bit more cache however. I also have a 50MHz Cyrix 486SRx2 that overclocks to 66MHz, but unfortunately I can't make an exact comparison as the SX motherboard I have is quite old and uses a very different chipset. I will bench it when I get bored though.
My 386SX motherboard (AMI 386SX XT Series 20):

Anyway, I would bet that if you ran your DRx2-66 using an ISA VGA card, your doom score would be around 5300-5400 realtics, which only makes it slightly faster than the TI486SXL overall. Do you have the pipelining feature enabled using the Cyrix DRx2 utility? I've heard the pipelining option can speed things up a fair bit.