Hi Gliba,
seems confusing 🙂
Raster pyramids (see help) reflect the content of single raster files at reduced resolution. That also (usually) means: They show excactly the same content as the data - just in reduced resolution. They can be stored in the raster itself or as additional files. While they speed up use of single files at lower resolution, an "extent request" that requires the system to mosaic together many images, would still require to read all files and the mosaic them together. Thats why there are Overviews in MosaicDatasets. They (usually) are reduced resolution pre-mosaicked "aggregated" partial views - thats why it takes some time to generate them, but afterwards they will be faster since the original files wont have to be mosaicked (again and again) when request are made for an area covering more than one base-file. By the way: You might run into problems without overviews quickly - since the limit of Images mosaicked together per request (property of the MosaicDataset) might be met - and that makes a lot of sense performance-wise.
But keep in Mind: In Mosaics there are situations where you want to use pyramids at higher resolution (instead of overviews) or where you create overviews in scales where there actually would be pyramids present - question of content, flexibility and usage. You could also have Mosaics that do have many layers of data that represents different scales and resolutions, but no Overviews at all. ..
Whenever you do not need the flexibility of a mosaic dataset (at least not often) and want all processing to be done already at any scale you need, you create a cache. Potentially this includes raster processes, removes aerial overlap and tiles the data up into little "chunks" - dumb but fast 🙂
So very likely when you use MoaicDatasets you'll have both: Overviews and a Cache.
I leave the comments on Block tiles to more knowledgeable people in that field 🙂
