Advanced applications for Distributed Hash Tables (DHTs), such as Peer-to-Peer Information Retrieval, require a DHT to quickly and efficiently process a large number (in the order of millions) of requests. In this paper we study mechanisms to optimize the throughput of DHTs. Our goal is to maximize the number of route operations per peer per second a DHT can perform (given certain constraints on the lookup delay). Each peer receives congestion feedback from the DHT, which it uses to adjust its routing decisions. This way, peers can avoid routing through slow parts of the overlay network and hence increase the rate at which they insert new messages into the DHT.We provide a numerical analysis of congestion-aware routing in DHTs and show that considerable improvements in throughput are possible compared to DHTs with proximity neighbor selection and strictly greedy routing.