If all OPTIM jobs are failing for some reason then it is likely that the rsh command that starts the jobs on remote nodes will run out of sockets. Using a single rsh command together with cp via nfs to copy results back seems to have eliminated the previous problems that we sometimes saw with socket errors.
Rate constant calculations with any graph transformation variant can be very memory hungry. They will die with a segmentation fault if they run out of memory.