> The obvious likely failure case occurs when one have multiple disks
> with completely zeroed-out first blocks.

This is not a `failure case'.  Such a disk cannot be used by other
OSes, or booted from, so we don't need to know the BIOS geometry.

There are some issues with installing a new machine with a blank disk
from scratch, but I believe the `arbitrarily pick one and warn the
user' method (where `arbitrarily' might mean `pick the geometry for
first BIOS disk') would solve this problem in 99% of such cases -- and
always result in a bootable installation on the first disk.

The cases where this wouldn't work are a corner case of a corner case;
basically they are all cases where someone is building a machine from
parts and installing a system on a disk other than the first.  It's
reasonable to expect such users to actually read the manual.  (I'm
assuming this will be documented, of course!)

> Writing out a cookie in the
> boot block solves this problem robustly but is worrisome.

It's not just worrisome; it won't work.  Virus-detecting BIOSes will
always lose with this.

When I suggested using a cookie, I meant putting it in the label, when
the label is written, so that once we've touched the disk we can
always identify it uniquely.  This was mainly intended to solve the
`map BIOS boot device to NetBSD device' problem, and doesn't solve the
problem of getting the BIOS geometry for installation porpoises.