Question to the authors: Any reason why hippyvm is not included for comparison in the paper? It does usually outperform hhvm on those benchmarks (but not on real world use cases which is maybe a good reason to include real world use cases more into such papers).