Mobile Cloud Computing (MCC) enables smartphones to offload compute-intensive codes and data to clouds or cloudlets for energy conservation. Thus, MCC liberates smartphones from battery shortage and embraces more versatile mobile applications. Most pioneering MCC research work requires a consistent network performance for offloading. However, such consistency is challenged by frequent mobile user movements and unstable network quality, thereby resulting in a suboptimal offloading decision. To embrace network inconsistency, we propose ENDA, a three-tier architecture that leverages user track prediction, realtime network performance and server loads to optimize offloading decisions. On cloud tier, we first design a greedy searching algorithm to predict user track using historical user traces stored in database servers. We then design a cloud-enabled Wi-Fi access point (AP) selection scheme to find the most energy efficient AP for smartphone offloading. We evaluate the performance of ENDA through simulations under a real-world scenario. The results demonstrate that ENDA can generate offloading decisions with optimized energy efficiency, desirable response time, and potential adaptability to a variety of scenarios. ENDA outperforms existing offloading techniques that do not consider user mobility and server workload balance management.