The simplicity and low-overhead of random walks have made them a popular querying mechanism for Wireless Sensor Networks. However, most of the related work is of theoretical nature and present two important limitations. First, they are mainly based on simple random walks, where at each step, the next hop is selected uniformly at random among neighbors. This mechanism permits analytical tractability but wastes energy by unnecessarily visiting neighbors that have been visited before. Second, the studies usually assume static graphs which do not consider the impact of link dynamics on the temporal variation of neighborhoods. In this work we evaluate the querying performance of Non-Revisiting Random Walks (NRWs). At each step, NRWs avoid re-visiting neighbors by selecting the next hop randomly among the neighbors with the minimum number of visits. We evaluated Pull-only and Pull-Push queries with NRWs in two ways: (i) on a test-bed with 102 tmotes and (ii) on a simulation environment considering link unreliability and asymmetry. Our main results show that non-revisiting random walks significantly improve upon simple random walks in terms of querying cost and load balancing, and that the push-pull mechanism is more efficient than the push-only for query resolution.