One last posting, and then I will be out of thoughts...
JDW, I know you were posting your experiences there, but the wording makes it sound a bit like the version of the Turbo040 board is speed dependent.
My understanding, which could be wrong of course, is that it is a time dependency. The early version of the Turbo040 is the one with many GAL PLDs on board and the FastCache socket on the back. The later manufactured Turbo040s are the ones with the large square ASIC and the cache soldered down on the board.
I spoke with a Daystar tech way back when about this once, IIRC. He claimed that the ASIC version was also faster. I guess (he didn't put it in these words) that they were able to tighten up some of the timings when they put it all in one ASIC rather than spreading it across PLDs. I guess that would make sense. They could probably do a larger combinational cloud between register steps in the ASIC and therefore eliminate some layers of flip flops.