iterative decoder (20) having a plurality of servers (22) that perform iterative decoding of a data block each, an input buffer memory (23) and a control unit (21) which performs a statistical multiplexing of data input, which is first stored in the input buffer (23) and subsequently processed by one of the servers (22). The input buffer (23) has N + L memory locations, where N and the number of servers (22) and L is the number of additional locations called. Each block to be decoded is received while all servers (22) are occupied is stored in one of The additional slots may be available, or it is lost if the input buffer (23) is completely filled. The number of additional locations and the number N of servers are calculated on the basis of a queue model of type D / G / N / N + L so that the probability P that a block _B or lost checks the condition P _B ≤α · IRON *, where * is FER the error rate on accepted blocks and α <1; α is typically of the order of 0.01. Method of manufacturing such a iterative decoder comprising a stage of design and material manufacturing step.