What I'm wondering about is whether to back-patch this. It's possible that people have written patterns like this and not realized that they aren't doing quite what's expected. Getting a failure instead might not be desirable in a minor release. On the other hand, wrong answers are wrong answers.
I'd vote to back-patch this. The unscientific reason on my end is that I suspect very few patterns in the wild would be affected and furthermore any using such patterns is likely to be in a position to change it match the existing behavior by replace the "(\1)" with the corresponding "(\w)" as you used in you example. We should probably suggest just that in the release notes. It is not a strongly held position and my first reaction was that introducing an error should be avoided. But regular expressions are tricky enough to get right when the engine does what you tell it...