The World Wide Web (WWW) has caught people's attention since its inception and today is a major source of traffic in the Internet. WWW traffic is typically asymmetric: servers receive small requests but answer with large number of packets. To ensure good service, WWW servers require a large egress bandwidth to the Internet. We propose an innovative approach to connect Web servers to the Internet via many outgoing connections. We employ cognitive packet networks to make packets learn which link provides the best quality for each client so as to balance the WWW traffic effectively. Egress bandwidth can easily be increased by increasing the number of links to the Internet. The paper describes our traffic engineering approach, an implementation model, a testbed platform in the context of the Linux kernel, and measurement data; these indicate the usefulness of our proposal.