Pericom was the maker of those bus switches. I don't know why I was thinking about Precicontact, the connector mfg, but I apparently merged the connector maker with the bus-switch maker to come up with my resulting error. Those switches are fast enough, contributing just a couple or three ns of delay, but add nothing to the output drive of an MCU, so they won't work with the older parts that drive only 2 mA or so Iol/Ioh.

My point, of course, was that being able to roll-yer-own is pretty powerful in the event you're able to capitalize on someone else's generosity in putting their work out there in the public domain. It enables one to "fix" what bothers 'em about that work, but it also carries with it a burden of understanding quite fully how that core the other guy built actually works. I don't know where the notion that those Dallas-designed MCU's have a long pipeline comes from, since they don't have a lot of clocks by means of which they'd operate one. It might be true of the multi-clockers, but the one-clockers can't be supporting a long pipeline in order to get the sort of performance that they do get.

I'll have to spend a little time with the Dallas parts and a logic analyzer just to satisfy my curiosity. I don't know when I'll get to that, however. It might be more productive to build a test fixture that feeds the DS89C4x0's pretty much every instruction that it can handle, place it in XRAM, map it as program space, and run it there as part of a do this N times loop, sampling, all the while, just how long it takes as compared to an 803x running it in external memory. I'll try to come up with something to tie numbers to this. I haven't seen any thorough exercise/evaluation results ... yet.

RE