Distributed real-time avionics networks have been subjected to a great evolution in terms of functionality and complexity. A direct consequence of this evolution is a continual growth of data exchange. AFDX standardised as ARINC 664 is chosen as the backbone network for those distributed real-time avionics networks as it offers high throughput and does not require global clock synchronisation. For certification reasons and engineering research, a deterministic upper bound of the end-to-end transmission delay for packets of each flow should be guaranteed. The Forward Approach (FA) is proposed for the computation of the worst-case end-to-end transmission delay. This approach iteratively estimates the maximum backlog (amount of the pending packets) in each visited switch along the transmission path, and the worst-case end-to-end transmission delay can be computed. Although it is pessimistic (overestimated), the Forward Approach can provide a tighter upper bound of the end-to-end transmission delay while considering the serialisation effect. Recently, our research finds the computation of the serialisation effect might induce an optimistic (underestimated) upper bound. In this paper, we analyse the pessimism in the Forward Approach and the optimism induced by the computation of the serialisation effect, and then we provide a new computation of the serialisation effect. We compare this new computation with the original one, the experiments show that the new computation reduces the optimism and the upper bound of the end-to-end transmission delay can be computed more accurately.