The main reference point for this is of course CDS
http://cdsweb.u-strasbg.fr/doc/UCD.htx but you can find some AstroGrid
documents on UCDs
http://wiki.astrogrid.org/bin/view/Astrogrid/OntologyDocs
One obvious starting point is what we have learnt from the AVO First Light Demo. My first thought is that it is very useful when you can answer a query by a partial UCD, e.g. plotting anything with a REDSHIFT whether it is REDSHIFT_PHOT or REDSHIFT_HC etc; or PHOT_*_V where * can be JHN or COUS or MAG or .... (e.g. in Aladin, for filters or for the SED tool..)
However you also need to be able to track how much uncertainty you are introducing (as well as using the more exact UCD if you want of course) - e.g. any old PHOT_MAG_V might do as part of a radio-to-x-ray SED but not for fine inter-optical comparisons.
So any restructuring of UCDs should take this into account.
cheers
Anita