In this paper, we study the performance of wireless networks with randomly located access points (APs) and randomly located energy-harvesting user equipments (UE) focusing on network throughput and system energy efficiency. We consider harvest-and-transmit protocol for wireless-powered UEs that do not have embedded power sources, but harvest energy from densely deployed APs. We first characterize outage probabilities and achievable data rates as a function of the system parameters, i.e., uplink and downlink operating intervals, using tools from stochastic geometry. Then, we formulate an optimization problem that maximize the system energy efficiency (measured by throughput per total consumed energy) while taking the feasible operation interval constraint into account. Despite the difficulty in obtaining closed-form expressions for the optimal solution, the problem can be solved using standard numerical tools. Simulation results demonstrate that broadcasting energy-bearing signal at higher power level improves energy efficiency. Furthermore, circuit power consumption is shown to be a key factor affecting the performance gain as well as optimal downlink/uplink time allocation strategy.