The problem with memory usage is lack of benchmarks. For most of the benchmarks that are on the page it'll be only the interpreter size. We're working on a solution (and on memory usage) though.