In a previous life I learned a number of lessons about network traffic and queue models. Basically, there are state changes and at these changes traffic can rise asymptotically. If you are managing a network you want to avoid these state changes and plan capacity accordingly. (see e.g. http://portal.acm.org/citation.cfm?id=1096231.1096237&coll=GUIDE&dl=ACM or http://www.tfhrc.gov/its/tft/chap9.pdf)
As we have re-launched our football site this fall we have developed a number of techniques to drive traffic and dramatically improved the content. Unfortunately, we didn't pay adequate attention to the performance capabilities of our server environment. As a consequence, we ended up with many more hits than we could service. As the site became overwhelmed the response time fell and the browser failed to load the page. So even though we received more "hits" (content requests) we generated fewer page views (completed requests).
In the physical world you can imagine a call center, for example a stock brokerage, where a sudden market panic generates a call volume that overwhelms the call center. Anxious consumers wait on hold for a short period of time but then hang up and re-dial, exacerbating the problem. Those that get through take more time because they don't want to be forced through the process of re-connecting to the call center. The call center receives more calls but services the same or fewer calls.
We're in the process of expanding capacity and improving our diagnostics to help us forecast demand. In the mean time we've lost traffic and probably alienated a number of users.
Traffic modeling and capacity planning are critical components of our business. Like all fundamentals, they need to be practiced every day.