Here is a quick analysis of the locks hanging off serverStatus:

• recordStats requires a read lock (Client::ReadContext) for each db
• locks requires "DBLocksMap::ref r(dblocks);" (not write lock related)
• cursors requires client cursor lock
• network requires local lock for network counters
• repl requires lock on "ReplSetImpl"
  • master/slave: local db, and possibly queries all replication sources (other server oplog entries – first/last entry).

The biggest issue is the recordStats since that requires a read lock for each db.