In this paper, we provide theoretical and simulation-based study of the delivery delay performance of a number of existing throughput-optimal coding schemes and use the results to design a new dynamic rate adaptation scheme that achieves improved overall throughput-delay performance. Under a baseline rate control scheme, the receivers' delay performance is examined. Based on their Markov states, the knowledge difference between the sender and receiver, three distinct methods for packet delivery are identified: zero state, leader state, and coefficient-based delivery. We provide analyses of each of these and show that, in many cases, zero state delivery alone presents a tractable approximation of the expected packet delivery behavior. Interestingly, while coefficient-based delivery has so far been treated as a secondary effect in the literature, we find that the choice of coefficients is extremely important in determining the delay, and a well-chosen encoding scheme can, in fact, contribute a significant improvement to the delivery delay. Based on our delivery delay model, we develop a dynamic rate adaptation scheme that uses performance prediction models to determine the sender transmission rate. Surprisingly, taking this approach leads us to the simple conclusion that the sender should regulate its addition rate based on the total number of undelivered packets stored at the receivers. We show that despite its simplicity, our proposed dynamic rate adaptation scheme results in noticeably improved throughput-delay performance over existing schemes in the literature.Australian Research Council (Discovery Project DP120100160)United States. Air Force Office of Scientific Research (Award FA9550-13-1-0023)