Computational models of a large metabolic system can be assembled from modules that represent a biological function emerging from interaction of a small subset of molecules. A "skeleton model" is tested here for a module that regulates the first phase of dynamic adaptation of oxidative phosphorylation (OxPhos) to demand in cardiomyocytes. The model contains only diffusion, mitochondrial outer membrane (MOM) permeation and two isoforms of creatine kinase (CK), in cytosol and mitochondrial intermembrane space (IMS) respectively. The communication with two neighboring modules occurs via stimulation of mitochondrial ATP production by ADP and inorganic phosphate from the IMS, and via time-varying cytosolic ATP hydrolysis during contraction. Assuming normal cytosolic diffusion and high MOM permeability for ADP, the response time of OxPhos (t_mito; generalized time constant) to steps in cardiac pacing rate is predicted to be 2.4 s. In contrast, with low MOM permeability, t_mito is predicted to be 15 s. An optimized MOM permeability of 21 µm s^-1 gives t_mito = 3.7 s, in agreement with experiments on rabbit heart with blocked glycolytic ATP synthesis. The model correctly predicts a lower t_mito if CK activity is reduced by 98%. Among others, the following predictions result from the model analysis: 1. CK activity buffers large ADP oscillations 2. ATP production is pulsatile in beating heart, although it adapts slowly to demand with "time constant" ~14 heartbeats 3. if the muscle isoform of CK is overexpressed, OxPhos reacts slower to changing workload 4. if mitochondrial CK is overexpressed, OxPhos reacts faster.