I'd love for someone to do the empirical test to see whether booting from the end-of-the-chain external Sony drive works or not, but it may well not. (I honestly just don't know.) Perhaps it said so in boldface letters in the original manual, but the first attempt at Googling for a copy of the HD-20's user manual came up empty. The only hit I found was this page, where the person does say:
"Although you can daisy chain another external 400K drive off the HD20, you still need to boot from the internal drive. Which stinks if your internal drive is out of order."
They then attempt to anyway, because the internal drive in their 512k is shot, but coincidentally their 512k decides right at that moment to blow up completely so there's no resolution of the question.
Even if it doesn't work then being able to do HD-20 *and* 400k floppy emulation would still be useful assuming you can make your internal drive work, IE, if you have a 512k with no boot floppies you simply boot from a floppy image with the device in floppy mode, format a real disk and drag a system over, and from that point use it to boot with the HD-20 emulation. And of course the other option is to toss Plus ROMs into the machine and solve the problem that way. (Yes, then it's not completely original, but if you intend to actually use the machine it's a solution that's easily reversed.)
Ah well. I've discussed this in some earlier threads, but something I've mused about tackling if I thought I had even the *slightest* chance of writing the device driver (either as a custom INIT or patching it into a custom ROM) would be interfacing a CF card to a Mac via the ROM sockets. Floating around out in Internetland there's an article/schematic describing an early Pre-Plus SCSI adapter that was interfaced via that route, and there were also Mac Plus ROM-compatible SCSI cards interfaced the same way. (The Mac Plus' SCSI implementation is *very* similar to the ROM socket hack. Eerily so, actually.) A CF card in minimal 8-bit mode has very similar interface requirements to the NCR 5380 and the command sets for dealing with the block device are broadly similar. (Assuming you use LBA and not CHS mode on the CF card.) The interface would only need three or four chips on a daughterboard and someone with actual programming talent could probably make that go pretty easily...
(On my "to do" list next time I'm fiddling with repairing my not-perfect Commodore PETs is hanging a CF card off its expansion bus and seeing if I can make that work in some useful fashion. The thing is, of course, a brain-dead driver for a PET is a lot simpler than a Mac OS driver... or at least it seems that way to me. I looked at the driver code for that ROM-socket SCSI card and just sort of glazed over.)