I definitely like this idea, except for the "non-keyboard specific part".
I think it wouldn't be too hard to do, at least at a basic level, for keyboards. As soarer pointed out though, that would likely lose some special features a keyboard might have. And as I mentioned in the other thread, Hasu's ps2 converter actually has this functionality already, and is pretty straightforward to modify.
Figuring out the USB<->USB aspect would likely be the most difficult.
And I assume this would have to be driverless, otherwise it sort of defeats itself, yes? I've had some decent luck with intercept drivers before, passing keyboard input to an external device and receiving it back in some transmogrified form, but that definitely requires drivers, so not plug and play.
Another thought: what about getting something like Synergy running an arduino or teensy, but having it do the translation prior to kicking it out over the network interface? You can pass network data over a USB port without drivers, but it can be a bit...wonky. And it would require that you run synergy on the host machine, but that can be done without an install, from a flash drive, for instance.
These are just thoughts, by the way, but I would be interested in pursuing this project. While for my current purposes, Hasu's project worked perfectly (post-modification), some day I may not have PS/2 keyboards, so having a project I can turn to in the future would be nice. So count me in on this one. I can do some of the software side, I would think.
Do we want to pool our resources and snag a QIDO, see if it can be reverse-engineered (outside the US, of course)? I don't have the hardware to dump and/or flash it, but someone here might.
Edit: Oh, and can we have the option for more than 104 key keyboards, please? Specifically, the terminal emulators which send unique codes. It would be nice if we didn't lose that ability. The USB HID spec has codes for those extra keys, after all, so why not just include them as a pass through type deal.