Re: Luck, or Unicode? Neither - just accurate coding to the API.
Well, the code tests the platform on which it running. So, if it is running on an already infected machine, and that machine is a Windows 9x box, the infection will spread no further.
Stuxnet, we are told, is a very complex piece of software, and the claim that this would have led to a BSoD on Windows 9x (an OS routine that hooks into CPU-level faults - e.g., double faults) suggests that it is low-level code (i.e. Kernel mode on an NT-based system). My expectation would be that the code probably wouldn't run at all on anything but an NT kernel, so the issue would never arise anyway on a 9x box.
If you haven't used Windows since the early 90s, you may not be aware that 'Windows' confusingly refers to three distinct and very different operating systems (we won't mention Windows CE). It is a very common and understandable confusion amongst Unix and Linux developers. The long-obsolete 9x and NT families share similar 32-bit APIs and could run a fair number of common 32-bit applications back in the 1990s. However, they have (wildly) different kernel architectures written by entirely different teams. David Cutler, who originally wrote VMS and went on to lead the NT effort at Microsoft never worked on 9x, to the best of my knowledge. The Stuxnet code tests the platform to make sure it is running on just one of those families - NT - which has evolved into the only Windows codebase Microsoft now maintains and evolves.