Also, I think I've figured out what the mysterious header bytes mean.
If your source was converted straight from a properly zipped ePub, then you get 53524353000000100000003000000001. If it came from any combination of un-prepackaged html/opf, it'll be 53524353000000100000002f00000001. If it's a no-source-files-added mobi to begin with, then the header bytes are 46434953000000140000001000000002.
And it seems that even the samples offered for the newer books at Amazon nowadays include the bloat (but only from the mobi conversion and cut off appropriately at the sample length), which looks like it's a useless expenditure to me.
Ah well, if they want to waste their server bandwidth for no good reason, that's entirely up to them. As long as they don't go back to charging that extra $2 Whispernet surcharge that they finally got rid of for Canadians.