These kind of sweeping statements are usually pretty obviously false. You can write anything cleanly at scale. Those MS-DOS sources that were released a while back are pleasant enough to read despite being in assembly.
The implied meaning of "impossible" in this kind of context is "unreasonably difficult". Just as it would be unreasonably difficult to write modern Windows in assembly, it would be unreasonably difficult to maintain a large CSS codebase without nested selectors.
You make a good point. CSS to SCSS doesn't seem like that big of a step up though. I don't really think nesting is really the greatest solution to the problem. It's got priority caveats over BEM, and can cause messes of it's own when used without precision.
