Fine-Tuning Vision-Language-Action Models: Optimizing Speed and Success