Bone density and other body composition phenotypes are complex traits with many contributing genetic factors, both shared and distinct. Genetic interactions are often suspected to contribute to the genetic architecture of such phenotypes, but understanding their role and extent is commonly limited by sample size. To reduce this limitation, we created a large F2 intercross of over 2000 B6.lit/lit x C3H.lit/lit mice to investigate interactions influencing femoral density, femoral circumference, body fat, and body weight. We used a combined analysis of pleiotropy and epistasis to infer how multiple quantitative trait loci (QTL), sex, and circulating growth factor IGF1 interact to influence all traits simultaneously. A large connected network with dozens of directed genetic interactions among multiple distinct QTL was obtained and analyzed, from which we obtained insights into the genetic architecture of this intercross populations. Recurrent patterns of QTL interactions with sex and circulating IGF1 suggested regulatory and compensatory roles for candidate genes. The QTL network was dominated by genetic interactions that reduced the occurrence of extreme phenotypes when the two interacting loci shared a common parental genotype. Rather than widespread genetic buffering, in which extreme phenotypes arise from strong synergistic interactions between QTL pairs, we found a genetic architecture characterized by (1) weak interactions that adjusted additive variance, and (2) genetic redundancy. This first class of phenotype-stabilizing interactions has effects that are close to additive and are thus difficult to detect except in very large sample sizes. We interpret these findings as a genetic mechanism of homeostasis for each inbred parental strain that is disrupted by allelic assortment via intercrossing. Our results suggest that while epistasis is often weak and unlikely to account for a large proportion of heritable variance, even small-effect genetic interactions can facilitate hypotheses of underlying biology in well-powered studies.