Seemingly, always forwarding requests could yield the most benefit for saving backbone bandwith. However, that is not always true. For a very hot page, forwarding control messages repeatedly may add up to higher bandwidth consumption than obtaining the page once. Also, each time a request is forwarded, the client will experience extra traversing delay from one proxy to another, if the second proxy is not much closer to the client. In contrast, after fetching the page, all future accesses would be satisfied by the receiving proxy locally. Moreover, request forwarding prevents hot pages from being replicated in multiple proxies; this may cause those proxies which keep on accepting forwarded requests for hot pages to become overloaded.
The main goal of the forwarding mechanism is therefore to select out those rarely referenced pages for bandwidth reduction. Distinguishing between ``hot'' pages and ``cold'' pages can be achieved by maintaining a reference history of each document on every proxy. A forwarding threshold can be used to decide if a document is a ``hot'' page. Requests for seldom referenced pages are forwarded, but those for ``hot' pages are filtered out for fetching.