Video simulcasting enables a sender to generate multiple streams for the same video program but at different rates, targeting receivers of different capacities. Each stream can then be distributed to the target receivers via broadcast or multicast channels. One of the critical issues in video simulcasting is to strike a balance between bandwidth consumption and the user satisfaction. This is because stream replication introduces redundancy which can potentially lead to excessive use of the bandwidth, while on the other hand there usually is a mismatch between a receiver's capacity and the limited choice of the stream bandwidths. We present a formal study on the above issue, which we refer as the optimal stream replication problem. We first formulate the optimal replication problem for a given session bandwidth budget and number of streams, with the objective of minimizing the expected mismatch for all the receivers in the session. A proper choice for the number of the replicated streams is critical in the overall bandwidth provisioning. Though it has been largely assumed that the number of streams is predetermined by service providers, we find that dynamic adjustment of stream number is possible in advanced video stream systems, such as the MPEG-4 Delivery Multimedia Integration Framework (DIMF), and systems employing fast video transcoding algorithms. Therefore, we also study the case of a flexible stream number, with the objective of jointly optimize the stream number and bandwidth allocated to each stream.