Patient-derived xenograft (PDX) models have been rediscovered as meaningful research tool. By using severely immunodeficient mice, high-engraftment rates can be theoretically achieved, permitting clinical stratification strategies. Apart from engraftment efficacy, tolerability towards certain cytostatic drugs varies among individual mouse strains thus impeding large-scale screenings. Here, we aimed at optimizing an in vivo treatment schedule using the widely applied cytostatic drug 5-fluoruracil (5-FU) for exemplary response prediction in colorectal cancer (CRC) PDX models. Four different individual CRC PDX models were engrafted into NOD.Cg-PrkdcscidIl2rgtm1Wjl (NSG) mice. Mice with established PDX were allocated to different treatment groups, receiving 5-FU, the oral prodrug Capecitabine, or 5-FU/leucovorin (LV) at different doses. Body weight, tumor size, and general behavior were assessed during therapy. Ex vivo analyses were done from blood samples, liver, as well as tumor resection specimen. Engraftment efficacy was high as expected in NSG mice, yielding stable PDX growth for therapy stratification. However, overall tolerability towards 5-FU was unexpectedly low, whereas the prodrug Capecitabine as well as the combination of 5-FU/LV at low doses were well tolerated. Accompanying plasma level determination of DYPD, the rate-limiting enzyme for 5-FU-mediated toxicity, revealed reduced activity in NSG mice compared with other common laboratory mouse strains, offering a likely explanation for the drug incompatibility. Also, the De Ritis quotient was highly elevated in treated mice, reflecting overall organ injury even at low doses. Summarizing these findings, NSG mice are ideal hosts for in vivo engraftment studies. However, the complex immunodeficiency reduces tolerance to certain drugs, thus making those mice especially sensitive. Consequently, such dose finding and tolerance tests constitute a necessity for similar cancer precision medicine approaches.