On Fedora Core 6 with 2 dual-core Xeons and 4GB of memory (yes, stacksize is unlimited). NetCDF 3.6.2. MPICH21.0.5p4.
1) Trivial, but both gfortran (4.1.2) and Intel Fortran builds with OpenMP are reporting "Initial basin volumes: TotVolume" to have a different value than is reported for runs using SERIAL or MPI and the same compiler.
2) Intel Fortran (10.0.023) builds of BENCHMARK using OpenMP continue to SegFault on benchmark1 with 2x2 tiling; 4x4 tiling finishes without issue.
For reference, fastest runs are with Intel F90 OpenMP. All runs that finish do so with correct KE,PE and TE, and Volume calcs by time-step are consistent.
g95 seemed to offer no major benefits and had issues with MPI. I abandoned it.