Comprehensive VLBI and multi-waveband monitoring indicate that a single superluminal knot can cause a number of gamma-ray flares at different locations. However, the often very rapid variability timescale is a challenge to theoretical models when a given flare (perhaps the majority of those observed) is inferred from observations to lie near the 43 GHz core, parsecs from the central engine. We present some relevant observational results, using the BL Lac object AO 0235+164 as an example. We propose a turbulent cell model leading to a frequency-dependent filling factor of the emission region. This feature of the model can provide a solution to the timescale dilemma and other characteristics of blazar emission.