Well, you really don't need to be a good engineer, but the skills required to do anything meaningful in theoretical computer science PhD program are more hardcore and rare than good SW engineers - from what I see, most of the people who can do theoretical CS are already doing it; and most engineering-oriented CS researchers couldn't/wouldn't cross to theoretical CS no matter what.
I mentioned only some subset of PhD applicants, which might not be small. So there are many other types of applicants who apply with different and various reasons.
As for your question, no, I don't think that scientists in all of the branches of CS need to be good engineers. But that is just my personal opinion.
I'm not sure how these students are filtered out at other places, but in Brown's CS department, PhD students are required to pass a "programming comp" that involves a fairly intense week-long programming project. You don't need to be a superstar coder, but you do need to be able to program effectively in order to pass.
It completely depends on what area you're going into. Software engineering, systems, and AI students obviously need significant coding experience. HCI and theory (ie math), not as much.
Right, but what I wanted to highlight - a person might be able to code quick sort algorithm or breadth first search, but unable to get paid doing real world coding - when it is the client or employer who needs to be satisfied.
So basically when you have a choice of being unemployed vs doing PhD and getting some studentship, you obviously prefer the latter path.
And I believe there are quite many people who can "code" but unable to get paid.
If you can "code" so poorly that you're unable to get paid, I can't imagine how you'd get even close to getting into a PhD program, even if your specialization is math of finite state automata or human interface guidelines.
Coding (or being DBA or analyst, depending on your area) to the level of getting an okay salary is quite trivial compared to that - from the PhD students I've seen the 'most unemployable' ones would be so not because of their ability to code but because of social/mental issues.
Theory cs could care less about coding abilities. It's never a bad thing to know how to code, but I can go though the entire PhD program without coding a single line of code. (Well not really, one of the professors in my PhD program committee recommended me to do one systems course. other than that, I just prove theorems. )