In an HSDPA system, multiple users are scheduled in a time slot due to constraints on the user terminals with respect to how many codes a particular user can utilize. We model the allocation of the codes at the so-called flow level. This results in a particular multiserver queuing model, where codes correspond to servers and multiple servers are allocated per flow subject to constraints on the maximum number of codes. In this context, we focus on minimizing the mean flow delays by utilizing flowlevel information on the remaining service times. While SRPT is the optimal policy for minimizing the mean delay in an M/G/1 queue, no such optimality results exist for the dynamic setting in multiserver models. We derive a heuristic SRPT policy for the system and evaluate its performance against the fair round-robin policy, which can be modeled at the flow-level as a processor sharing system. The results demonstrate that using SRPT-like scheduling can significantly decrease the overall mean delays, as well as the conditional delays.