Now that I think about it, one could use a small javascript library embedded in the index page to make a number of additional requests and interpret them as the

Now that I think about it, one could use a small javascript library embedded in the index page to make a number of additional requests and interpret them as the correct types via data: URLs. That would be a lot of messy hacking to shave off a few hundred ms, but might be an interesting exercise to undertake...