To accommodate the dynamic connectivity and networking condition in DTN, most existing routing schemes require inferring future contact opportunities to select message relays. However, such relay only has an incidental effect on helping opportunistic data delivery. In order to again a further understanding of the intrinsic uncertainty of relay efficiency, this paper engages in an empirical investigation on the relay selection schemes. Firstly, it introduces a more effective opportunistic relay selection strategy based on the estimation of residual message delay, which utilizes pairwise contact records and the elapsed time from last contact. Next, based on the underlying opportunistic vehicular network extracted from large-scale realistic vehicle traces collected from urban areas, it investigates the efficiency of relay selection with an empirical view. The questions this study focuses on are: What is the probability that a selected relay can make the end-to-end delay reduced? How much the latency can be saved by a properly selected relay? Such empirical study is signficant for the protocol design and application deployments in the future opportunities vehicular networks.