Semantic validation of the effectiveness of a schema matching system is traditionally performed by comparing system-generated mappings with those of human evaluators. The human effort required for validation quickly becomes huge in large scale environments. The performance of a matching system, however, is not solely determined by the quality of the mappings, but also by the efficiency with which it can produce them. Improving efficiency quickly leads to a trade-off between efficiency and effectiveness. Establishing or obtaining a large test collection for measuring this trade-off is often a severe obstacle. In this paper, we present a technique for determining lower and upper bounds for effectiveness measures for a certain class of schema matching system improvements in order to lower the required validation effort. Effectiveness bounds for a matching system improvement are solely derived from a comparison of answer sets of the improved and original matching system. The technique was developed in the context of improving efficiency in XML schema matching, but we believe it to be more generically applicable in other retrieval systems facing scalability problems.