Four user-level threads packages are compared in this paper. It presents a quick, but complete description of the general threads model, and describes how fine-grain parallelism may be applied to the thread model.
The authors discuss, for each package, the application programming interface (API) model, implementation, and the pros and cons. The authors did benchmarking tests on the implementations, and comment on the results obtained.
The paper focuses on implementations of user-level threads, which limits the domain of application (all benchmarks come from the numerical analysis domain), and which might be too restrictive for readers interested in a general-purpose multithreads implementation.
The bibliography provides a lot of useful of pointers, and the authors are certainly authorities in their field.