Thanks for the links on PCI-X. I'll take a look at them.

As far as the separate bus thing, I tend to agree with you. But the consumers tend to push the market in this area, and really, really fast graphics cards are the things to have these days, and they do push a lot of bandwidth, especially when you consider all the texture maps and whatnot that they're pushing around. And separating that off from the rest of the data being pushed around is actually an advantage for both the graphics cards and the rest of the devices. Of course, this has been attempted before in the name of VESA Local Bus, and that fell off the planet in short order. AGP may go the same route, but change is less likely to happen these days, IMO, because of the much larger, less knowlegeable group of people that own computers these days. Of course, since most of those people just swap computers out on the whole, that may not be an issue.

I'd also like to point out that there is a specification for a separate networking bus that can accomodate modems and NICs. CNR, maybe? Of course, no one uses it, because the manufacturers see no advantage in it, and they're right, so the consumers don't push it. (Edit: In rereading your post, I notice that you talk about CMR. Maybe you typo'd CNR? If so, I didn't know that it was just attached to the PCI bus. What possible reason is there in having a different connector?)