This paper compares two path selection strategies for Optical Burst Switching networks: a) the path from source to destination is chosen randomly from the set of the k shortest paths; b) the path selection is based on a probabilistic model for the link demands. In the second strategy, the set of shortest paths from source to destination is ranked according to the probability of their overall link demands. The path selection schemes are evaluated for a specific scenario of Grid over OBS networks. Our results show that, in general, the probabilistic model outperforms the random k shortest path routing scheme, this improvement is relevant for small job sizes.