There's no good evidence behind that number, but that doesn't mean it's not reasonable. Furthermore, that number will likely vary for different areas of physics.
The reality is that no one really does a good job of keeping track of what percentage of physicists do what largely due to poor methodology and the cost of tracking human beings. The guy who got a high profile professorship is in the university and AIP stats, but the dude who disappeared to work at Starbucks sometimes is and sometimes isn't. It only takes a small percentage of that bottom tail to disappear and your expected values all shift to the right (E(X) | X > y). In this case, the problem is that you don't know if they left because they couldn't find a job or found another one they liked more, or both, or neither. Even if they do tell you, interpreting their response can be a pain.
However there are lots of less formal ways of getting a feel for the difficulty level. Look at polls that ask physics grad students whether they'd like to stay in academia and compare it with the typical number of position openings, for instance.