Embodiments are directed to a method of regulating client agent backup jobs in a backup server by defining a server parallelism value specifying a maximum number of active backup jobs that can be simultaneously performed by the backup server, maintaining a plurality of queues in memory and including an active queue storing active backup jobs comprising a number of current backup jobs not exceeding the server parallelism value, and one or more overflow queues storing backup jobs exceeding the server parallelism value, receiving backup job requests from a plurality of clients and processing the received backup job requests through the plurality of queues, and transmitting a hold command to the plurality of clients if the plurality of queues are full. The overflow queues may comprise a wait queue and a sleep queue.