Hi,
I have not searched or found the link, but would like to give a very short answer to the general topic to clarify (hopefully):
PYRAMIDS are reduced resolution representations of SINGLE IMAGERY FILES. The main intention of pyramids is allow you to "see" this image in a broader scale range (and read less data for a given "scale").
OVERVIEWS are reduced resolution images that combine the area of potentially many images. They allow to "see" an area without the need to read and combine the data of potentially many images. The main intention of overviews is to take away the need to mosaic many images together for requests in large scales every time and to read less pixels in general. Reduced processing, bandwidth need, ...
Your question "do I need PYRAMIDS, OVERVIEWS or both" can not be answered in general for any usecase. It will depend on analytical needs, processing applied, data.
For optimized performance especially of Services, reducing the amount of file-reads and pixels to be processed is key. The best way to do this is OVERVIEWS. In many cases having NO PYRAMIDS but OVERVIEWS will be the most efficient solution. In cases where you need to identify single image source in a broader scale range, want to benefit from optimized resampling at any scale, have say frequent changes in base data and probably use a completely different dataset instead of overviews, pyramids might be useful, needed, good. So - it depends
Regards
Guenter