well your output is very puzzling. We do this all the time, although I have not yet tried 2.4.25. I think the last time I did this was 2.4.22
It's almost as though something in the kernel is trashing the ramdisk when it starts up.
But you say it did work from cdrom, which makes this even weirder.
I hope it is not some random DMA from ethernet landing sometime after kernel starts, but it sounds too reproducible.
I think you're stuck with some more etherboot debugging. Do you have to use net boot?
ron