IMHO the CASS radical set would be the one to use for this. It has the support of a major national body and is used in several dictionaries, including Xiandai Hanyu Cidian and John DeFrancis' ABC Dictionary.
That's what I was thinking, although the lates ABC dictionary, the Comprehensive, no long uses the CASS table but KangXi with extensions to handle simplified characters. The order of the CASS table used in ABC also differs from the most recent editions of the Xinhua Zidian and Xiandai Hanyu Cidian. However, I think once characters are mapped to the 189 CASS radicals, switching around the order would be easy.
I expect too that generating a CASS radical index could be done automatically using the kRSKangXi field in Unihan and the mapping table between CASS and KangXi radical numbers in DeFrancis - kRSCASS would make sense.
I've been working on a program like you describe. I think it will handle the majority of characters, but in a significant percentage of cases the mapping is not direct and will need to be hand-corrected.
Are all of the CASS radicals themselves encoded in Unicode?
I'm not sure and I'm not an expert, but from my experience it seems so.
Erik
