The problem with a single cache proxy is that it is both a bottleneck (poor scalability) and a single point of failure (lack of robustness). There is a trade-off between scalability and latency: the larger the number of clients, the more likely to get a cache hit, but the larger the number of clients per server, the larger the latency. Speaking about robustness, it's clear that if the proxy cache fails, there is no other possibility than to get the documents directly from the primary servers.
Fortunately a single solution has been proposed that would achieve both scalability and robustness: to have multiple independent caching servers cooperate. There have been two basic architectures proposed to address this issue. The following two subsections give an outline of each architecture.