Second, the author use autocannon which is a Node.js based to loadtest uWebSocket which made the result to be very skewed because it's not fast enough to outperform C even with a HTTP overhead
Which means that even their own result isn't even accurate on its own