Magma mixing and recharge is an ubiquitous process in the shallow crust, which can trigger eruption and cause magma hybridization. Phenocrysts in mixed magmas are recorders for magma mixing and can be studied by in- situ techniques and analyses of bulk mineral separates. To better understand if micro-textural and compositional information reflects local or reservoir-scale events, a physical model for gathering and dispersal of crystals is necessary. We present the results of a combined geochemical and fluid dynamical study of magma mixing processes at Volcan Quizapu, Chile; two large (1846/47 AD and 1932 AD) dacitic eruptions from the same vent area were triggered by andesitic recharge magma and show various degrees of magma mixing. Employing a multiphase numerical fluid dynamic model, we simulated a simple mixing process of vesiculated mafic magma intruded into a crystal-bearing silicic reservoir. This unstable condition leads to overturn and mixing. In a second step we use the velocity field obtained to calculate the flow path of 5000 crystals randomly distributed over the entire system. Those particles mimic the phenocryst response to the convective motion. There is little local relative motion between silicate liquid and crystals due to the high viscosity of the melts and the rapid overturn rate of the system. Of special interest is the crystal dispersal and gathering, which is quantified by comparing the distance at the beginning and end of the simulation for all particle pairs that are initially closer than a length scale chosen between 1 and 10 m. At the start of the simulation, both the resident and new intruding (mafic) magmas have a unique particle population. Depending on the Reynolds number (Re) and the chosen characteristic length scale of different phenocryst-pairs, we statistically describe the heterogeneity of crystal populations on the thin section scale. For large Re (approx. 25) and a short characteristic length scale of particle-pairs, heterogeneity of particle populations is large. After one overturn event, even the "thin section scale" can contain phenocrysts that derive from the entire magmatic system. We combine these results with time scale information from U-series plagioclase age dating. Apparent crystal residence times from the most evolved and therefore least hybridized rocks for the 1846/47 and 1932 eruptions of Volcan Quizapu are about 5000 and about 3000 yrs, respectively. Based on whole rock chemistry as well as textural and crystal-chemical data, both eruptions tapped the same reservoir and therefore should record similar crystal residence times. Instead, the discordance of these two ages can be explained by magma mixing as modeled above, if some young plagioclase derived from the andesitic recharge magma which triggered the 1846/47 AD eruption got mixed into the dacite remaining in the reservoir after eruption, thus lowering the apparent crystal residence time for magma that was evacuated from the reservoir in 1932.