Just an idea I wanted to get down... If I understand correctly, we have to block the parser to wait for <script> and stylesheet loads to complete before we carry on parsing. This means we have to load those resources essentially one at a time. One conceptually straightforward way to deal with this would be, when the parser blocks, scan ahead through unparsed content looking for additional resource-loading markup, and prefetch those resources into the cache (in a way that means we don't have to revalidate them if they actually do get loaded once we unblock the parser). The scan doesn't have to be 100% reliable; if it loads a resource we end up not needing, we should be OK in theory and in practice.