By downscaling the dose by a factor of 33.3x, computing the histogram, and then upscaling the result, you essentially scale up histogram quantization error by 33.3x.
To maintain error at the same level, despite of this aggressive scaling, you need to set a smaller histogram bin width (=DVH step size). You can get current step size by opening the Python console (Ctrl-3) and typing:
slicer.modules.dosevolumehistogram.logic().GetStepSize()
To set a new value, type for example this:
slicer.modules.dosevolumehistogram.logic().SetStepSize(0.006)