The spec says Chrome is right (was recently changed). Chrome is not the only browser. iOS browsers, the Android browser and Chrome for Android have always used the new behavior. The reason for the change is faster scroll performance, particularly on mobile.
Really? Can you point to the spec change? Did I miss a working group decision on this somewhere? Last I checked, the discussion on this got nowhere, with WebKit folks claiming this change was needed to optimize stuff and Opera and Mozilla folks claiming that optimizing scrolling worked just fine without this behavior change...
The CSS 2.1 spec was last updated in 2011 and does not include any mention of this new stacking context behavior (which was only proposed on the CSS mailing list in May 2012 and has not yet made it into any spec).
No, they changed it to no longer comply with that spec, actually. Please do read the spec, not assume it says what the "html5rocks" marketing folks want it to say.