Monaco is a native-code, shared-memory multiprocessor testbed implementation of flat committed-choice languages such as Flat Guarded Horn Clauses. A reduced abstract-machine instruction set facilitates decision-graph code generation and traditional optimizations based on dataflow analysis. We describe the compilation process and present empirical measurements characterizing performance gains enabled by various compilation and runtime kernel construction techniques. Monaco ran 1.6-4 times faster than comparable parallel systems, yet achieved slightly better multiprocessor speedups (10.7 on 16 Symmetry processors). Among the compiler optimizations, common subexpression elimination achieved the largest speedup of 15%.