Metacomputing aims at using the computing power of computers linked by networks for various applications, which used to be tackled by dedicated supercomputers a few years ago. Web-based metacomputing offers a cheap and efficient solution for parallel and distributed computing by exploiting idle cycles from the Internet. Research literatures have been published to report the state of the art of metacomputing by addressing the developed prototypes architecture, support methodologies and technologies, programming APIs, example benchmarks and experimental performances. This work tries to answer the question: why the metacomputing system is undesirable in performance, applicability and reliability. In this work, the system's components are modelled as three roles: client, worker and manager. Their performances are individually investigated in terms of the theory of constraint: objectives, requirements, prerequisites, conflicts and injections for identifying core problems and exploring efficient solutions. The main contribution of This work is to be a very first literature as conceptual guidelines for design and optimisation of metacomputing systems.