Networks of constrained devices play an important role in the Internet of Things (IoT). In such networks congestion may lead to significant performance decrease and is a recurring phenomenon given the restricted hardware capacities of constrained devices and the limitations of low-power radios. The Constrained Application Protocol (CoAP) designed for IoT communications defines a basic congestion control mechanism for the exchange of messages with end-to-end reliability between two endpoints. Yet, for transmissions without end-to-end reliability the CoAP base specification does not determine any congestion control mechanism. Proposals for congestion control mechanisms for unreliable CoAP communications are made in the CoAP observe and CoAP Simple Congestion Control/Advanced (CoCoA) Internet drafts, which introduce a static and a dynamic rate limitation for outgoing messages, respectively. In this paper experimental evaluations are carried out to determine the performance of the different proposals. We find out that, in contrast with the other analyzed approaches, CoCoA is able to maintain high performance in all the analyzed scenarios, thanks to its adaptive nature.