Yea it is kind of intentional. There's some cases where there's pretty large response bodies and I figured it's better to write to disk for those cases, then to run out of memory. Thankfully most really large responses are already streams so they don't need to go through this process.