problem with performance of simple unconventional job

I created a simple 'solver' (a sh script) that actually just extracts metadata information from a benchmark and writes it to stdout. It does not matter whether the benchmark is pronounced as successful or failing. I did manage to create a couple jobs and start them. However, the script runs very quickly, but the job is executing very slowly - a 2-4 orders of magnitude more slowly than I expect. I can't see a way to tell how loaded the cluster is the cluster status tab shows nothing useful). Furthermore the status update on the job progress page is not updated in any consistent manner.

SMT/david_cok/Info ...