This paper presents an empirical comparison of the Semi-Naive, Generalized Magic Sets, Generalized Counting, and Separable query evaluation strategies as applied to queries on multi-rule recursions. I used each of the methods to evaluate queries over randomly generated relations. For each query there is a critical density range. If the base relations are below the density range, Generalized Magic Sets, Generalized Counting, and Separable provide roughly equal performance and are significantly better than Semi-Naive. Above the density range, Generalized Magic Sets degrades to Semi-Naive, Generalized Counting is much worse that Semi-Naive, while Separable is still significantly better than Semi-Naive. This suggests that special purpose algorithms such as Separable can greatly improve the performance of a recursive query processor.