Hmmm. That's a 1 out of 9 failure rate on the chips so far. That's not so good. I hope you just happened across the bad apple and that's not typical.
Have you tried running something like NewerTech's RAMometer memory test (also available in Guages) or some of the memory tests in TechToolPro? Check the chips bitwise...
Fabulously excellent work.
Hmmm, browsed the thread again, rereading your progress.
I also suggest you test RAM performance (RAMometer will give you a MB/S tested indicator) with and without your DTACK wire connected. Is it possible that without DTACK, the CPU is timing out and assuming the data is on the bus at some point, but that with DTACK, the transaction might complete several cycles sooner?
If there's some logical reason why the above paragraph is misdirected, feel free to ignore it.