There must be something peculiar that these machines have in common. I mean, they work with a regular USB keyboard, right?
Baffles me, because I am sending a basic, boot-compatible report, at all times!
The PS/2 to USB adapter is interesting, because it too will be a composite device. Perhaps the BIOS deals badly with all composite devices in some way, and has only been tested with the most basic of keyboards? It doesn't seem likely, but...
When Windows boots, it should be completely resetting all of the USB devices. I am at a loss to imagine what the BIOS can be doing to prevent this. Same presumably applies to any other OS these days.
I'm hoping that if this problem becomes more common a pattern will emerge that makes clear what is going on. It would certainly be useful to test as many composite devices as possible, to help in that.