Ok, here's a summary of what I understand:

 + We _will_ need MOP support of some kind for those models that flat
   out refuse to boot a TFTP image of any size.

 + TFTP will be easier _in_general_, because almost everything has
   TFTP support but not everything has (or will have) MOP support.

 + A two stage TFTP loader will benefit those models that can TFTP boot,
   but pork out at around 1MB.

 + The mopd patches I wrote used ECOFF because at the time ECOFF was
   easier than ELF (for me).  An ECOFF kernel is pretty much an a.out
   kernel with an extra header or two.  An ELF back-end should be easy
   to do.  I've _never_ used BFD - this may make life easier...

 + From an Ultrix support POV - just use an elf2ecoff'd kernel with
   the normal Ultrix mop support.  For a 5000/150 here that just
   can't tftp boot, I just

	addnode -h 08-00-2b-30-17-8f -l fredri /dlboot0/fredrick/netbsd

   One advantage of Ultrix's mop node database is that the image
   can live anywhere on the system - it doesn't need to be located
   underneath /tftpboot.

Is this a reasonable summary of the state of the world?

Simon.