Energy harvesting is an important feature that can be implemented in mobile devices to provide them with extended autonomy, yet it poses several challenges in terms of optimal battery usage. In this paper, we highlight that some non-ideal effects, especially the so-called "charge recovery," can have a dramatic impact on the operation policy of autonomous devices. To do so, we construct a Markov model, where we introduce a bidimensional battery value, including the apparent energy level, which is what available at the electrodes to power the device, and the actual energy level stored in the battery. When power is not drained by the device, these levels tend to equalize due to charge recovery, but, in the case of intense battery usage, they can be significantly different. We show that this non-ideality leads to considerably different estimates of undesirable events such as battery outages, and may cause a general underutilization of the devices if not properly accounted for.