Normally when a browser asks for Content-Encoding: gzip, the http server needs to schlep the file into memory, compress it, and only then is it able to send the output to the client. Since redbean reads files from a zip archive, that file is already compressed. So what it does is it uses mmap() to load the zip executable in memory. Therefore instead of read()+write() it can just call writev() on the mapped memory area. That enables us to skip the extra copy. The kernel can just copy the gzipped file contents directly to the network wire without having to pay any userspace context switching penalties.
