The popularity of Energy Harvesting Devices (EHDs) has grown in the past few years, thanks to their capability of prolonging the network lifetime. In reality, EHDs are affected by several inefficiencies, e.g., energy leakage, battery degradation or storage losses. In this work we consider an energy harvesting transmitter with storage inefficiencies. In particular, we assume that when new energy has to be stored in the battery, part of this is wasted and the losses depend upon the current state of charge of the device. This is a practical realistic assumption, e.g., for a capacitor, that changes the structure of the optimal transmission policy. We analyze the throughput maximization problem with a dynamic programming approach and prove that, given the battery status and the channel gain, the optimal transmission policy is deterministic. We derive numerical results for the energy losses in a capacitor and show the presence of a \emph{loop effect} that degrades the system performance if the optimal policy is not considered.