HI Karl,
Thanks for your interest in the NE64. I think that you are going to have some problems implementing a Ethernet bootloader.
Sure, the flash on a NE64 can be programmed in situ (but with the code running from RAM) however, you're going to need some form of stack running, to be able to download the code. Problem is, the stack will be larger than the RAM on the NE64. You're gonna need to download a page at a time (512 bytes, as I recall), cache that in RAM, then jump to RAM and write the new code, but this is all hideous. As you said, how do you ensure that you don't erase/program something that you need to go back to.
A more elegant solution would be with the ColdFire MCF52235 device, with 256kbytes of flash, arranged as 2 banks. You can run code from one flash bank and reprogram the other, and the InterNiche Niche_Lite stack easily fits into one bank. That's just what we did with the Ethernet Seminar lab material back in 2Q06 - the stack lived in one bank and the webpages in the dynamic flash file system live in the other bank.