I assume you are looking at the table on slide 7. I think that is the percentage of all who completed relative to those who started it, so whether many CS undergrads go intro industry (which is most likely the case) or not is irrelevant. However, I can offer anecdotes from my experience (recent PhD in CS). First, it is usually easy to find good jobs in the industry while you are doing your PhD. I think half of students in my department that did machine learning and computer vision didn't finish their PhD and went off to work at Google and the likes. It is very tempting, because the hours in a PhD program are long, stress is high and money is never good. And many CS PhD end up working in the industry anyway. Second, writing a lot of code isn't necessary if you're doing pure theory --- but even then you might be making small prototypes or using proof assistants, so a no-programming research in CS is rare. Whether you need to master many new skills depends on your background and what you want to do for your dissertation.
To clarify on going to industry... If you are a top student with a bachelors in a field like Physics or Biology - there's aren't as many post-undergrad options as the top CS student. Therefore, the top Physics and Biology majors are more likely to head to grad school. Those top students are more likely to finish on time.
Again - just a hypothesis.
I think your point about mid-Phd CS folks having options is stronger. If you're 2 years into a Biology Phd program, your options are limited. If you're 2 years into a CS Phd program, the sky is the limit.
I hear you on CS theory. My impression during undergrad was that the grads were doing a lot of programming in the coursework, and many beyond.
