Behaviometrics in multi-factor authentication schemes continuously assess behavior patterns of a subject to recognize and verify his identity. In this work we challenge the practical feasibility and the resilience of accelerometer-based gait analysis as a behaviometric under sensor displacement conditions. To improve misauthentication resistance, we present and evaluate a solution using multiple accelerometers on 7 positions on the body during different activities and compare the effectiveness with Gradient-Boosted Trees classification. From a security point of view, we investigate the feasibility of zero and non-zero effort attacks on gait analysis as a behaviometric. Our experimental results with data from 12 individuals show an improvement in terms of EER with about 2% (from 5% down to 3%), with an increased resilience against observation attacks. When trained to defend against such attacks, we observe no decrease in classification performance.