This is a classic case of localization, which can occur when materials or components lose strength. This phenomenon is well described in the literature and easily observed in nature. It is not always easy to make the behavior in a model follow that in nature, but the cause is the same.
If you look at the resultant force in the model, it is uniform, or very nearly so, throughout the loading. The stresses in the layers are also uniform until the concrete reaches cracking stress. It is not expected that the entire structure will lose strength uniformly. In fact, when cracking occurs in a real concrete structure, you see distinct localized cracks. The whole wall does not simultaneously disintegrate in tension. In the physical case, where the cracks occur, the steel is carrying the entire load. Where the wall has not cracked, the concrete shares the load with the steel.
This is also what is happening here. The top row of elements has cracked, and the steel is carrying all the load. For the rest of the rows of elements, the steel and concrete are sharing the load. Just before cracking, the stress is uniform. However, it is not perfectly so because of numerical round-off in the calculations. In a real structure imperfections would have the same effect. Once one region starts to crack, it relieves the stress on the rest of the region and it does not crack. That is called localization.
Figure 1 - Localization of material nonlinearity
Another physical example of localization is the necking seen in a steel tensile specimen. During loading, the strain is uniform over the central region of the specimen. Once strength loss occurs, it localizes and most of the straining occurs in a local region, the “neck”. The length of this region is controlled by the geometry of properties of the specimen, and is typically on the order of the width of the specimen. A finite element model of this behavior will also show localization, but the necking length will be controlled by the mesh size. This is not physically realistic, so care must be used to choose an appropriate mesh size.
For the cracking problem, the calculated crack size will also be equal to the mesh size. The user needs to decide what a reasonable size is for this. If the goal is detailed stress modeling, the physical behavior will need to be considered to decide the mesh size. If the goal is practical design information, such as for performance-based design, detailed models are not warranted and may even be misleading. In the latter case, use the largest elements possible. The strain demand will be averaged over the larger element and will be more useful.