Ha. I can see your skeptical.
Seems to be working fine. I loaded up Photoshop and 3dsmax at the same time, had both of them processing. Wintune.. Windows also gobbles everything up as a disk cache. Was not able to do this with the 3 chip ones, I could sometimes boot windows, and as you say, it would then fall over. I haven't finished testing, it could still be an issue, but it seems very solid so far.
There is a problem at 50Mhz.. It does fail.
I have mix of 60ns and 70ns in one bank. It fails at 50mhz. 40Mhz everything is rock solid and passes just fine. 50Mhz, flakes out booting windows. So now I just need a pair of nice 60ns 4mb simms. But I might as well buy another 4 x 4mb of the same. Shipping is killing me to Australia, every american seems to overpack. But I now know 9 chip works. I am pissed that the 32mb of brand new 60ns 3 chip parity didn't work and that I eventually will end up buying 32mb in 9 pin parity in multiple shippings.
In the end I will have to sell off a lot of stuff that I have purchased but not using in this build.
In then end I want all 32mb at 55mhz. By my reckon, I think that might just pip the faster pga132 boards (at 40 or 50mhz). I am looking at writing some code to enable the 256Kb cache in the chipset as well, which I think is possible and will make up for some of the slower performance.
Eventually I will capture the video (vga-composite-captured composite) and put together a little video on the build and the machine in action. Playing Jazz jack rabbit II, Red alert, C&C, Simcity 2000, Doom, 3DSMax, Photoshop, Vistapro3, win 95 benchmarks. It is quite a useful little machine. It feels very much like a mid spec 486 in its usability than what most people remember a 386 is like. I can see why people are pretty skeptical.
If you ever get a chance to experiment with a peak Dm board, try the bios I provided, see if it helps with higher clocks and full memory. The same motherboard with the Abit bios I tried, was not stable at 50mhz and I couldn't get 32mb stable either, but I didn't play with all the options. I guess Abit had tightened the timings (possibly non user editable ones) which causes problems (but would make the board faster at the same speed).
I will also try to write a program tho write those registers to enable the 256kb cache.