The documentation on locking during a database repair is ambiguous.
The command documentation implies a global lock is obtained. This alone is unclear because it's not obvious if this was updated after per-database locks were implemented.
To clear up confusion, a user may find themselves on this page, which very strongly implies that repairing a database will only hold the lock on the database being repaired, not the "database" as in the whole process.
I just attempted to repair a database on a 2.6 replica set secondary and all operations across all databases were blocked.