We have just published a new paper with a case study of the Intel MIC architecture.
This is a simplified CFD application solving 2-dimensional equations of shallow water flow with a memory bandwidth-bound stencil operator. It is expressed in hybrid OpenMP+MPI framework, and runs natively only on the Xeon Phi architecture. With coprocessors, we had to tune more carefully than with CPUs. However, only one line of 10+ year old code had to be changed to achieve good performance and scalability.
The source code and sample running script can be downloaded at the same page.