Conclusions
Expecting to be able to produce a good, parallel implementation using OpenMP, with very little effort, is, in general, unrealistic. (Nevertheless the authors are optimistic that, in this case, good performance is achievable with not too much more effort .)
The OpenMP version is certainly easier to write, but the extensive use of Fortran 90 array features, which is ideal for these codes, is not yet handled well. (However, it is believed that the next release of OpenMP will include the parallelisation of Fortran 90 intrinsics, which should certainly help.)