Embodiments of the present invention are directed to systems and methods for queuing and sending messages to recipients according to historical latency values associated with each recipient. In some embodiments, a plurality of messages are received, each message including a network address of a recipient. The messages are sent to the recipients by threads that remain active (i.e., cannot be used to send another message) until confirmation responses are received from the recipients. Latency times are measured between when the messages were sent and when the confirmation responses were received. The latency times may be used to assign future messages to queues designated by certain latency ranges.