Normally you visualize imagery at the highest possible level of detail using a desktop application. It is standard to have reduced resolution dataset in a web viewer (for resources allocation and performances optimization). Uploading smaller portions of your orthomosaic, you could get the highest level of detail possible.
The ortho that is generated during processing is not the ortho that is displayed on the Cloud. The ortho is processed and optimized for viewing in a web browser through a tiling process. The resolution of the Cloud orthomosaic is determined by the level of detail (LOD) that tiles are generated for. The LOD that is produced is not always going to be the highest possible given the resolution of the original ortho. The reason behind is that this is an exponentially computationally intensive process and in order to maximize the functionally of the Cloud as well as ensure optimal disk storage procedures, it is necessary to limit these operations.
I hope I answered your question.