BitTorrent is currently a very popular protocol for P2P file sharing, and most of BitTorrent users access the Internet from residential networks. In this paper we compare the legacy BitTorrent protocol with EE-BitTorrent, a proxy-based version recently proposed for energy efficiency, in a residential scenario. We show that the performance achieved by users is strongly influenced by the uplink throughput allowed by the access network. When the available uplink rate is low, the legacy BitTorrent protocol performs poorly and EE-BitTorrent outperforms it, in terms of average download time and energy consumption at the user's PC. The opposite occurs when the uplink rate is good. Motivated by these results, we designed and implemented AdaBT, an adaptive algorithm that dynamically selects the most efficient BitTorrent option (i.e., legacy or proxy-based), depending on the operating conditions experienced by the user. Our experimental results show that AdaBT is able to reduce significantly the download time provided by either the legacy BitTorrent or EE-BitTorrent