There are at least two downsides to storing all of the convert.conf data in memory.  First, it's wasteful, when many (most?) libraries require none, or very little of it.  Second, it's only loaded at server startup, meaning that to change a transcoding option requires a complete server restart.  If stored in the database and only retrieved when a track requires transcoding, then a web interface (native or plugin) could be written to allow editing of the conversion strings and they could be used without restarting the server.  The shipped text conver.conf file could still be kept for "resetting to defaults" if such ability were desired.  As non-track information, this would be additional data that would need to persist a clear/rescan type of library scan.