My colleague ran some benchmarks on our installation of I-TASSER at the NIH Biowulf cluster:
https://hpc.nih.gov/apps/I-TASSER.html#bench
We would appreciate if you took a look to make sure that our observations and conclusions there are valid. Our installation is modified to copy and extract the I-TASSER database to local disk before starting the run in order to avoid heavy I/O over the network and improve run-time.