Both can be used to max out whatever your most limiting resource is.
http://blog.kubernetes.io/2016/03/1000-nodes-and-beyond-upda...
is the load over 9000?
Does it open a single TCP connection, then fire all load testing requests over that connection, or does it open multiple TCP connections and spread the load testing requests over multiple connections?
It's a bit unfortunate that because of the decision of the browser vendors to support only HTTP/2 over TLS (for valid reasons) now a lot of libraries and tools support only that mode. For lots of other applications HTTP/2 without TLS would be helpful too.