Scholars paid attention to the relationship between homework frequency and academic performance, however, fewer of them noticed that the relationship might be nonlinear and may vary across different types of homework. This study aims at exploring the nonlinear relationship between the different types of homework frequency and mathematical academic performance. To reach this goal, the study utilized the Multilevel Piecewise Regression Model (MPRM) and educational assessment data with 11,007 8th-graders students from a city in China. The results emphasize the importance of considering the nonlinear relationship between homework frequency and academic performance, as well as the differing effects of various homework types. Specifically, a higher frequency of Practice Homework (PH) seems to be positively related to academic performance, overwhelming Simulated Test Homework (STH) seems to be inefficient and overwhelming Extension Homework (EH) or Integration Homework (IH) has no positive effect on academic performance. According to the results, educators should not assign excessive STH (no more than once a month) to avoid hurting students' academic performance, while maintaining a higher frequency of PH may be more appropriate.