During parallel mongorestore jobs, a worker routine will block while it waits for a createIndexes command to return. This can take a very long time for many-gigabyte collections. We should explore the performance hit we would get by allowing a new collection to restore while we are still waiting for index builds from another collection.

I'd wager we could get some real throughput increase by allowing a couple of index builds to happen as we pick up a new collection to restore. This will likely vary by storage engine, but it is worth exploring.