Prey diversity is thought to mediate the strength of top-down and bottom-up effects, but few experiments directly test this hypothesis. I assembled food webs of bacteria and bacterivorous protist prey in laboratory microcosms with all combinations of five productivity levels, two top predator treatments (present or absent), and three prey compositions. Depauperate food chains contained one of two edible prey species, while more diverse food webs contained both edible prey species plus two additional less-edible/inedible prey. Equilibrium theory predicts that prey diversity should weaken the top-down and bottom-up effects on trophic level biomasses, due to density compensation among prey species. Top-down effects should increase with productivity in food chains, but decrease with productivity in food webs. Results revealed highly dynamic top-down effects, the strength of which varied more over time than among treatments. Further, top-down effects did not merely vary in absolute strength over time, but also in relative strength across different prey compositions and productivity levels. It might be expected that equilibrium models would qualitatively reproduce time-averaged results. However, time-averaged data largely failed to support equilibrium predictions. This failure may reflect strong temporal variability in treatment effects combined with nonlinear density dependence of species’ per-capita growth rates. Strong temporal variability in the strength of top-down effects has not previously been demonstrated, but likely is common in nature as well.