Right now probably nowhere. I have seen no place like that and the fact that the code works more or less kind of prooves that. But we're only depending on compiler behavior to hope to overwrite whichever register it loads into! All it takes is someone in the future to shorten a loop or add a spinwait (we already have some of those in CHDK) and you'll have a problem which will cost hours for someone to debug. I can see no point in your argumentation. However unlikely it may sound for you, why not just play on the safe side, espcially if we both agree that it means no harm performance-wise? Even if it works now, you're just playing with fire. Yes, I admit I'm paranoid but what is your reason for deliberately walking the path of undefined behavior?
If you really hate that patch that much, just ignore it, I'm not going to be offended or anything. I'm just trying to convice you or trying to make you convince me