I agree that in retrospect, it would have been better to have the key id be
the high-order bits of the fingerprint rather than the low-order bits. I,
too, think this was a mistake. However, that decision was made before this
working group was formed. (And before I joined PGP, Inc. for that matter.)
I believe that changing this and making it be dependent on the algorithm
type would be utterly wrong. It would add one more little gnarly bit into a
system that is already filled with too many gnarly bits.
If this blunder needs to be fixed, the correct way to fix it is to make a
V5 key structure. (For that matter, there are a couple other things I'd
fix, too, if we made a V5 key structure.) However, I think that as
unfortunate as this is, simplicity and compatibility override aesthetics,
and we should just live with things as they are.
Jon