@DiapDealer: From a (very) quick glance at the plugin code, KindleGen doesn't really enter the equation, the plugin's using the dualmetafix python script to add a 12 character one from a cropped random uuid (I'm guessing to more closely match the length of a real ASIN) to the metadata of the KG output.
Since it's mobi header metadata, KU properly preserves that when splitting. So, all's well

.
It's just that the option is disabled by default, and one may not be aware that trying to use such KF8 files directly will risk upsetting the Kindle's framework

.
And while doing it by default makes sense for Calibre, I'd argue that @Doitsu's choice was the right one here? He documented most of the underlying issues fairly well, and even I had forgotten how deeply messy it could be, so ;p. (That, or it got 'worse' since FW 5.5.x. But I'm way too lazy to check on an older device ^^).