There is vastly more music released, that could be part of an online music marketplace, than any central system can track.
I think a robust system would combine multiple central registries with overlapping data, and some very Internet-like distributed identifier mechanism, e.g., DNS-based URIs (aka web URLs) that individuals can create themselves to identify their works.
For example, a central service could create a digital works registry that takes URLs as input. Those URLs could then be used as IDs embedded in media files, etc. And, the central service could be queried with other identifying info, e.g., a sound fingerprint, to retrieve the URLs when they get misplaced.