Late to the party again, but....
One of many things I'm confused about is this:
Does this phenomenon apply only to Type1 fonts, or all fonts (ttf, otf, ...)?
And if the former, does it apply to otf fonts that were converted from Type1 (as the GhandiSans example would suggest)?
Also do I understand correctly that the phenomenon is observed whether you upload an epub (sigil compressed vs. WinRar compressed), OR whether you upload a mobi file built locally via kindlegen (which version? does it matter?) from the corresponding S or WR epub source. Even though both mobi files are properly functional
before upload to DTP?
A comment:
I'm also interested in the experiment where the WR epub is compressed using "defX" instead of "defN" compression, and whether it can be replicated on linux using the standard zip program (at various compression levels).
Then there are the "extra field" and "text vs binary" issues. The combinations and permutations are overwhelming.
Albert