CVR does not modify the source materials you provide in any way, so yes you can definitely make changes to download time by adjusting the sizes of scenes and objects within the scenes.
It would be hard to pin a generic “right size” for everything, as it very much depends on what the object itself represents, and what size within the scene the object will be displayed at.
For instance, the images that are built into CVR, such as the arrows and map pin, are around 200×200 pixels. That works because they are generally not stretched very large within the scene. But if there were a large image to be displayed that had fine detail, you would certainly want that at a higher resolution.
What you don’t need, and what I have seen in many scenarios, is full resolution photos from today’s phones or digital cameras to show a small thumbnail. You end up putting in very large files which do nothing but slow the download time of the content.