Horizontal and vertical heterogeneity as a result of size-structured processes are important factors influencing indirect effects in food webs. In a whole-lake experiment covering 5 years, we added the intermediate consumer roach (Rutilus rutilus) to two out of four lakes previously inhabited by the omnivorous top predator perch (Perca fluviatilis). We focused our study on the direct consumption effect of roach presence on zooplankton (and indirectly phytoplankton) versus the indirect effect of roach on zooplankton (and phytoplankton) mediated via effects on perch reproductive performance. The patterns in zooplankton and phytoplankton abundances were examined in relation to population density of roach and perch including young-of-the-year (YOY) perch in the light of non-equilibrium dynamics. The presence of roach resulted in changed seasonal dynamics of zooplankton with generally lower biomasses in May–June and higher biomasses in July–August in roach lakes compared to control lakes. Roach presence affected perch recruitment negatively and densities of YOY perch were on average higher in control lakes than in treatment lakes. In years when perch recruitment did not differ between lakes as a result of experimental addition of perch eggs, total zooplankton biomass was lower in treatment lakes than in control lakes. Phytoplankton biomass showed a tendency to increase in roach lakes compared to control lakes. Within treatment variation in response variables was related to differences in lake morphometry in treatment lakes. Analyses of the trophic dynamics of each lake separately showed strong cascading effects of both roach and YOY perch abundance on zooplankton and phytoplankton dynamics. Consideration of the long transients in the dynamics of top predators (fish) in aquatic systems that are related to their long life span involving ontogenetic niche shifts is essential for making relevant interpretations of experimental perturbations. This conclusion is further reinforced by the circumstance that the intrinsic dynamics of fish populations may in many cases involve high amplitude dynamics with long time lags.