While I pretty much agree with you that microbenchmarks are always suspect, it's also worth noting that HTTP2 will change the landscape a bit. All those bundles of icons in fonts and sprites can be loaded in parallel going forward...so, lots more very small files.
Further, I rarely worry much about web server performance, because I have so many other pieces in the stack that are the bottlenecks. In fact, almost everything in my web application stack is slower than my web server...so, I don't sweat the web server, except in cases where I can find a good argument that it matters. So, a benchmark of one very fast web server against another very fast web server (and most of the major players are very fast, at this point) is not going to convince me to change.
I find a performance-based sales pitch much less appealing than one that told me the primary focuses were security and predictability of configuration. Too many hours spent pulling my hair out with Nginx config
I guess it depends. For example, not if the Go interface would be used merely for the websocket part, while the raw content is still served by the C code alone.