As discussed...

When walking over a ~ 3M document collection the server is locking all collections for ~ 300-400 ms as it pulls more data from the server to the cursor.

While there are concurrency issues, it would be handy to pull less data (hence locking less).