> On the other hand, ethernet numbers are unique, which is rumoured, if
> I heard correctly, to be used by unscrupulous software suppliers and
> web administrators to log the browsing activities of unsuspecting
> visitors.

In point of fact, Microsoft's "globally unique IDentifier numbers" generated
by Windows contain the Ethernet address if a machine has an Ethernet card
(and the current time if a machine doesn't have one), plus some other bits
(boot time and a sequence number, I think).  These aren't *guaranteed* unique,
but it's a fairly good bet.  (Hence their desirability for unscrupulous
tracking; but they're also used for other purposes on Windows.)

> But without a registry, no uniqueness can be guaranteed :-(

The IEEE enforces the uniqueness of the first 3 bytes (the Organizationally
Unique Identifier, or OUI) and vendors are supposed to enforce the uniqueness
of the remaining 3 bytes.  Barring screwups, Ethernet (and token ring and
FDDI and HIPPI and ...) addresses can be considered unique.  (A side note:
if I remember correctly, AX.25 also uses 6-byte addresses generated from
an OUI issued for the purpose plus a hash of one's FCC issued amateur call
sign into 3 bytes.  Unfortunately, I know of at least one case where the
FCC issued the same callsign to two individuals!)

Of course, if you haven't got such a card, you're back to the question of
where you store such an address, as well as where you get it in the first
place (anyone up for a NetBSD-maintained OUI?).