This paper investigates L-system evolution through experiments with a simulation platform of virtual plants. The conducted simulations vary the occurrence probability of terminal versus non-terminal symbols and study its impact on the evolutionary performance of the system. The results reveal a variant of the exploration-exploitation dilemma. A closer look at individual runs allows to discover a range of emergent evolutionary dynamics. In particular, the activation and improvement of previously dormant production rules leads to variation in the fixation rate of mutations. The corresponding fitness leaps suggest that L-system evolution derives much of its creative power from the mobilization of randomly drifting non-addressed rules. The observed patterns are related to the phenomena of positive and negative selection, neutral mutations and junk DNA in the natural genome. Keywords virtual plants, L-systems, artificial evolution, artificial life