Complex computer simulations are a class of applications that demands high performance processing power in order to be realized in a feasible time. To achieve this processing power, networks composed of non-dedicated machines are increasingly being investigated. An efficient scheduling scheme is one of the most important issues to make a better use of these resources. In this paper we present an architecture for scheduling complex computer simulations aimed at heterogeneous non-dedicated machines which relies on information provided by the models that are being simulated. Furthermore, a case study demonstrates how the proposed architecture can assist in the execution of complex simulations applied to the protein structure prediction problem, which is one of the most important current challenges in structural bioinformatics.