The Qt 3 to Qt 4 transition was painful enough for too many projects to make similar breakage very unpalatable for the community. Sort of "been there, done that, didn't enjoy it" situation. And what you are suggesting (replace every container etc with a different one) would be even bigger breakage.
And of course from cost (be it money or time) point of view, touching basically everything in the Qt source code to strip out Qt classes would be gigantic work. Not going to happen, there's neither business case nor developer demand for breaking everything, so there's almost nobody who has any interest in making it actually happen.
Qt COW is thread safe, programmer does not need to worry about that. For performance critical parts, feel free to use something else, but for the usual GUI code, COW is very convenient. Also remember, Qt 5 is already so old that it couldn't require move semantics support from C++ toolchain, which is also a performance issue.
Moving forward, what I hope for future is, C++ enabling reflection and introspection (in ways required by Qt features) without extra code generators (like Qt's moc). That'd be great for Qt 6, if such a thing ever comes. Another thing which would greatly benefit Qt is C++ removing the need for manual implementation of the PIMPL idiom. And supporting C#-like extension methods might make transition to standard C++ containers more likely, with missing functionality added as extension methods (current need to mix methods and free functions is not very nice IMO).