Since SO_REUSEADDR is (and was for at least 9 years) set this is most likely not caused by existing connections in status TIME_WAIT.
See net.c: #ifndef WIN32 ss_opt = 1; /* Unimportant if this fails */ (void)setsockopt(sock, SOL_SOCKET, SO_REUSEADDR, &ss_opt, sizeof(ss_opt)); #endif So the likely cause is a premature start of the new process. Unfortunately I currently can not reproduce this.