On-demand routing protocols are widely used in mobile ad hoc networks due to their capability of adjusting to frequent network topology changes within acceptable routing overhead. In order to further reduce routing overhead, especially the overhead from the network-wide flooding in the route discovery phase, two techniques named route caching and searching localization are usually performed. In this paper, we reinvestigate these two techniques, in particular their joint effect on the routing overhead. For quantitative analysis purposes, we define one essential parameter for each technique: route caching validation probability and local searching radius. Based on the analytic results, we propose a new routing strategy that adapts to the current caching availability and is self-tunable towards the optimal performance. We demonstrate through extensive simulations that this routing strategy can reduce the routing overhead greatly under general scenarios.