To rephrase Idolse's answer in a different way, it is worth remembering that the Conversion pipeline in Calibre at core is a three stage process:
- Convert input into a common underlying representation (XHMTL) based
- Carry out standarised manuiplulations of the XHTML internal representation. This stage woul dbe working on flattened CSS, and possibly producing revised CSS.
- Convert XHTML based internal representation to a given output representation.
The phases are completely independent of each other. That is why the calibre plugin architecture works so well in providing great flexibility of conversion combinations as there are separate plugins for each input type and each output type. In fact there is no technical reason why you could not have a particular format with only an input or output type if the complementary plugin does not exist.
The fact that in an EPUB->EPUB conversion the input and output formats are the same does not change the overall process. Therefore the same manipulations of CSS and the like take place even if to a human perception they might be avoidable. The part that is doing that manipulation is not aware of the fact that the user selected the same input and output formats and that in this 'special' case that might not ALWAYS be necessary