This is not surprising because there is at least one artist actually called 'bach'

http://musicbrainz.org/artist/f71af296-6f77-48c9-97d8-2eb95472e9fa

and the sensible logic is always to prefer an exact match on name rather than alias,

But 99% of the time when someone enters 'Bach' they mean Johanne Sebastian Bach and we want Musicbrainz to work as well as it can for users. Previously we document boosted a hard-coded list of artists and labels to resolve this. Although this approach was not perfect because the list was incomplete it did solve the most common cases, this approach is not working since moving Lucene v3 to Lucene v4 because its creating ridicously large field norms for some reason so currently disabled

I cannot think of an alternative way to change the algorithm so that it would work for this case without breaking the logic for other cases. But maybe there is an alternative solution that does not need a hard-coded list, any ideas ?