Embankment dam safety depends on reliable foundation seepage control to prevent internal erosion and piping failures. Conventional designs rely on homogeneous or simplified layered models, which often mask the influence of realistic geological heterogeneity on seepage predictions. While extensive parametric studies exist on cutoff walls and impervious blankets, their performance under spatially variable, heterogeneous foundations remains inadequately quantified. In this study, two distinct heterogeneity representations are developed from 179 in-situ permeability tests: a layered single-pervious-zone model and a spatially detailed multi-zoned profile using the Closest Point Algorithm. Over 2,000 steady-state, 2D finite element simulations evaluated four broad configurations of partial cutoff walls, impervious blankets, and their combined systems. Results demonstrate that while total head dissipation is similar across models, exit gradients are highly sensitive to the heterogeneity representation. The detailed model predicted exit gradients 25- 30% higher at the critical core-toe interface. An upstream impervious blanket provided robust control, reducing the critical gradient by approximately 75%. For combined systems, placing a cutoff upstream of the blanket maintained safe gradients, while placing the cutoff downstream risked significant gradient amplification. Consequently, for deep heterogeneous foundations without bedrock, a full-length upstream blanket paired with an upstream cutoff is recommended, with final safety verification requiring spatially resolved heterogeneity models to avoid non-conservative design.