The planet candidates discovered by the Kepler mission provide a rich sample to constrain the architectures and relative inclinations of planetary systems within approximately 0.5 AU of their host stars. We use the triple-transit systems from the Kepler 16-months data as templates for physical triple-planet systems and perform synthetic transit observations. We find that all the Kepler triple-transit and double-transit systems can be produced from the triple-planet templates, given a low mutual inclination of around five degrees. Our analysis shows that the Kepler data contains a population of planets larger than four Earth radii in single-transit systems that can not arise from the triple-planet templates. We explore the hypothesis that high-mass counterparts of the triple-transit systems underwent dynamical instability to produce a population of massive double-planet systems of moderately high mutual inclination. We perform N-body simulations of mass-boosted triple-planet systems and observe how the systems heat up and lose planets, most frequently by planet-planet collisions, yielding transits in agreement with the large planets in the Kepler single-transit systems. The resulting population of massive double-planet systems can nevertheless not explain the additional excess of low-mass planets among the observed single-transit systems and the lack of gas-giant planets in double-transit and triple-transit systems. Planetary instability of systems of triple gas-giant planets can be behind part of the dichotomy between systems hosting one or more small planets and those hosting a single giant planet. The main part of the dichotomy, however, is more likely to have arisen already during planet formation when the formation, migration or scattering of a massive planet, triggered above a threshold metallicity, suppressed the formation of other planets in sub-AU orbits.