The need is to be able to
identifiy logical units (which I refer to as LUNs)
uniquely, no matter where they are and to whichever FC
network they might end up getting connected to at any
given point in time.

That's exactly what the SCSI device ID is for.

Hence anything that is not portable or not unique
world-wide is suspect. I feel the SCSI device ID falls
in this category.

The SCSI device ID, the NAA version in particular, is designed to be 
world-wide unique and portable (stays the same no matter where or how in 
the world you attach the device).

Since you're working with filesystems, another thing you should consider 
is a filesystem serial number.  The device ID has the problem that if you 
move your filesystem to another logical unit, its ID changes.  And if you 
replace the filesystem on a logical unit with another filesystem, the new 
filesystem has the same ID as the former one.  An ext3 filesystem has a 
UUID which is supposed to be worldwide unique identifier of the 
filesystem.  Your idea of placing your own ID file in the filesystem is 
similar.

Yet another approach would be to record the
host/bus/target/lun of each logical unit. However this
might not work in case of a re-arrangement of the
network or even re-seating HBAs inside the servers
themselves; am I right?

Yes, you should cross that one out.  It's only slightly better than using 
/dev/sdX.

So if a [WWNN] is indeed assigned to a single SCSI LUN,

It isn't.  It's assigned to the target.  Each target can have multiple 
logical units.  Each logical unit can have multiple LUNs.  And since it 
isn't a SCSI concept, there's no standard interface in Linux for querying 
it.  Various FC hardware and driver software have various ways to query 
such fibre channel info.