An SR (Selective-Repeat) ARQ (Automatic Repeat reQuest) protocol is used to recover from packet errors effectively over a low-quality communication channel. This SR ARQ has a problem of large delay due to resequencing of received packets. To mitigate this problem, the PFRS (Per-Flow ReSequencing) scheme was proposed, where the resequencing is performed independently for each upper-layer flow, while detection of lost packets and associated retransmissions are performed on the basis of the whole flows multiplexed over SR ARQ. This paper models the SR ARQ protocol, where the maximum number of retransmissions is limited, by a collection of simple stop-and-wait protocols, and shows numerical calculation results for the delay distribution of retransmission and resequencing. The validity of the analysis is confirmed by comparing numerical calculations with simulation results. The results prove the effectiveness of the PFRS scheme for the case where the number of flows over SR ARQ is large.