Programming languages tailored to particular groups of users can often be constructed by removing unwanted features from a general purpose language. This paper describes the use of simulation techniques to predict the savings in compilation cost achievable by such an approach. The results suggest a function which describes the effect of changes in the power of a language on the compilation cost of an algorithm expressed in that language: when features not actually used by the algorithm are removed from the language the cost of compiling the algorithm decreases moderately, but when features that are needed are removed, the compilation cost increases sharply