It has been well known that network coding can achieve better network throughput in certain topologies by allowing coding at intermediate nodes. However, the benefit of network coding in P2P content distribution is controversial in recent literatures. In this paper, we seek to investigate how network coding performs in a BitTorrent-like protocol from a practical perspective. To validate the benefit of network coding in BitTorrent, we first implement NCTorrent, a network coding enabled BitTorrent client. We then track about 10,000 real-world BitTorrent session's user behaviors and use the user behaviors to drive the experiments. Based on the results we establish, our conclusion with respective to the benefits of network coding is pessimistic. Network coding does not help that much in real world scenarios as claimed in previous work. In particular, the logged data also reveal that the last block problem, where network coding is believed to be helpful, only appears with very low probabilities.