Mitochondrial genome stability depends on purifying selection, yet this selection is fundamentally limited by the mixing of mitochondrial populations. Because selection requires variance, any process that homogenizes mitochondrial populations reduces its effectiveness. We introduce a coarse-grained parameter α describing effective mitochondrial mixing and show that selection efficiency scales with (1−α)2, implying a critical threshold above which purifying selection fails.This constraint resolves a key limitation of existing models, which assume that variancegenerated by the mitochondrial bottleneck is preserved. In reality, mitochondrial populationsare continuously redistributed by intracellular dynamics, including fusion—fission processes [9], which erode variance in heteroplasmy across cells. We further show that intracellular dynamics may favor mitochondrial variants with replication advantages but reduced energetic efficiency, creating a conflict between within-cell and between-cell selection. This conflict necessitates selection at the level of whole cells. We propose that mitochondrial quality control is a temporally structured, multilevel process. Oogenesis generates variance, early embryogenesis provides a window in which variance is preserved and fitness differences are expressed, and processes such as cell competition implement selection at the cellular level [2, 1]. Uniparental inheritance further contributes by suppressing mixing at fertilization [7, 6]. This framework predicts that increasing mitochondrial mixing or reducing variance impairs selection efficiency and provides a unified constraint-based explanation for mitochondrial genome stability.