Replication is one of the primary techniques used to improve the quality of distributed content service. It generally reduces user latencies and increases a site's availability. However, to our knowledge, there is no systematic framework that combines the structure of both content and service components of a Web application to design effective replica hosting architectures. Recent advances in interconnected and multiple content distribution network (CDN) architectures render this problem even more complex. In this study, we develop a systematic framework for designing and evaluating large-scale, component-based replication architectures for Web systems that are driven by both the quality and effectiveness of service provisioning on the service network. The proposed framework employs a combination of problem decomposition, configuration evaluation through controlled system simulations, and a neural-network-based feedback learning mechanism in the exploration of the design space. A case study demonstrates the viability of the framework. The framework can be an effective decision support tool for a system designer to systematically explore design options and select an appropriate design configuration that best meets the desired design objectives.