Dirk has the answer to Pea Genebank Database Hell:
A simple solution would be to agree on a set of prefixes that unequivocally identify each collection preserving those already widely accepted and then to clean up the passport data to link up potential duplicates. The 53674 Pisum accessions in Genesys could possibly be compacted to 1/10
Which is hopefully what DOI will do, though whether it will be possible to apply them retroactively, so to speak, is another question.