From what the
LinuxBIOS people have done in the past, I'd say it's not that much of an amazing achievement. Most of the speed increase seems to be due to them ripping out hardware detection and removing IDE support. Also it's 200ms to get the kernel loaded and running, it doesn't mean your application is actually running at this point.
Probing the IDE interface takes a few seconds and you have to wait for drives to spin up and initialise before they'll properly respond.
As it's an embedded platform they're targeting, then they can hardwire the configuration so they don't need to do any detection of available RAM, working out how fast the processor is etc...
Okay, you could do the last part on the empeg but you can't really take out the IDE interface code for obvious reasons