Wireless systems, and mobile ad hoc networks in particular, are more likely to experience transmission and routing errors than their wired counterpart. Factors like the lack of infrastructure, node mobility, and random radio link quality can contribute to significantly higher error rates in these networks. In addition, errors have a more serious impact on the network's resources, due to limitations in bandwidth and battery power inherent to the wireless ad hoc environment. This further complicates the task of designing scalable routing protocols, since larger networks are likely to experience even more errors, which may lead to slower convergence, longer end-to-end delay and unnacceptably high number of retransmissions. In this paper, we focus on the impact of error prevention and recovery on the scaling properties of on-demand protocols for ad hoc networks. Our analytical study, based on the evaluation of the Witness Aided Routing (WAR) and the Dynamic Source Routing (DSR) protocols, shows that the lack of localized intervention in handling errors translates eventually into lack of scalability, both in terms of performance and resource consumption. As route length increases, the performance of DSR degrades dramatically, especially in the presence of fluctuating wireless link quality. Even for small routes, DSR's lack of an error handling mechanism leads to very low probability of success when there is a non-zero probability that links are not bidirectional. On the other hand, WAR remains relatively insensitive both to the length of the route and to variations in mobility and call rates, and has a higher tolerance to radio link instability. This indicates that lacalized error correction can increase route effectiveness and alleviate the effects of short-lived radio link problems to an extent that allows the protocol to scale with the network size.