For best visual quality, render the volume image so that the size of a texel is about the size of a pixel. Besides making it easier to see density details in the image, larger images avoid the problems associated with under-sampling a minified volume.
Reducing the volume size will cause the texel data to be sampled to a smaller area. Since the over operator is non-linear, the shrunken data will interact with it to yield an image that is different, not just smaller. The minified image will have density artifacts that are not in the original volume data.
If a smaller image is desired, first render the image full size in the desired orientation, then shrink the resulting 2D image.