A system and method are provided for managing reader and writer threads in a caching proxy server. In general, a caching proxy server operates as an intermediary between a web server and a number of client devices. The clients send requests for digital assets hosted by the web server to the caching proxy server. For each request, or more particularly for each group of concurrent requests, for a particular digital asset, the caching proxy server operates in either a decoupled writer mode of operation or a reader/writer mode operation. In addition, while serving the requests, the proxy server may switch between the decoupled writer and the reader/writer modes of operation depending on one or more criteria.