We report on our findings after running a number of on-line performance monitoring experiments with a biomedical parallel application to investigate levels of performance at hardware resources distributed across a computational Grid network. We use on-line application monitoring for improved computational resource selection and application optimization. We used a number of user-defined performance metrics within the European CrossGrid Project's G-PM tool together with a blood flow simulation application based on the lattice Boltzmann method for fluid dynamics. We found that the performance results observed during our on-line experiments give us a more accurate view of computational resource status than the regular resource information provided by standard information services to resource brokers, and that on-line monitoring has good potential for optimizing our biomedical application for more efficient runs.