The defining property of two-state models of protein folding is that the measured relaxation rates are independent of the starting conditions and only depend on the final conditions. In this work we compare the kinetics of the very fast folding villin subdomain measured after a large change in denaturant concentration using an ultrarapid microfluidic mixer with the kinetics measured after a small temperature change in a laser T-jump experiment and find a significant difference in the observed folding kinetics. The final conditions of temperature and denaturant concentration and the use of tryptophan fluorescence as a probe are the same in both experiments, while the initial conditions are very different. The slower mixing kinetics show no evidence of the faster phase in T-jump experiments, which would support models of on- or off-pathway intermediates. Rather we interpret the combined mixer and T-jump experiments as evidence of an ensemble of unfolded states, some of which are traps. The ensemble after dilution from high denaturant is more expanded than the ensemble after an increase in temperature and, on average, takes longer to reach the native state.