That crystalizes my understanding of the situ, believe it or not.
Given that everything
appears to be working as intended until some point in the KDP process ... that means there has to be something quantifiably different about the mobi built from a normal (valid) Sigil epub, and the mobi built from the "hinky," ePub-ish WinRar beast. It's possible that some of the raw, debug output of KindleUnpack might shed some light on what that difference might be. There's got to be something there (or not there) that KDP notices and says; "this we like," or "this we don't like." It's either that, or you've truly stumbled across a loophole that allows you to achieve something that they have no intention of supporting any more (embedded Type1 fonts).
Thanks for the GhandiSans pointer. I'll poke around and see if there's anything obvious to infer from any differences in the mobis built using the two different methods, but I'm certainly no authority. I just have a decent working knowledge of the "guts" of a kindlebook and of KindleUnpack's (formerly MobiUnpack) codebase. I'm a noodler.