Parallelism versus memory allocation in pipelined router forwarding engines