Among other benefits, multithreaded systems allow the hiding of communication latency in distributed-memory systems. Then, though, the benefits of multithreading usually appear to be either ignored or overestimated. This is especially true when compared with approaches based on global structuring of communication. We therefore examine potential contexts for the use of multiple threads based on typical computation/communication patterns in both synthetic programs and real applications. We look at where multiple threads are beneficial or where structured communication may provide similar or even better performance. We also consider when and how much latency hiding can still be achieved if no special hardware support is available. Our experiments are conduced on the EARTH-MANNA system.