"Not trying to hire Linus" is a bit of a straw-man. There are a lot of folks that fall between "Linus" and "have to actively look for a job". Probably tens of thousands of people in the programming field.
The important point that I'm getting at, however, is that the dynamics of hiring shift dramatically as you move up the skill / connectedness scale. As you move towards the higher percentages potential employees are not competing for employers, but employers are competing for hires. And the dynamics of hiring are dramatically different at that point. If you're designing heuristics for hiring, you don't want to create a low-pass filter, and I believe that synthetic puzzles are exactly that.
The point of raganwald's post was on the surprisingly high cost of false-negatives. What I'm postulating is that your suggestion is just yet another way to reduce the false-positives rather than a way to reduce the false-negatives.
If it increased the number of overall respondents as well as the number of respondents who were above a certain bar (passed a phone screen), then even if it created other false negatives, it still seems to have helped.
Good developers don't just go around submitting their resumes to every job in existence, and not filtering out false negatives was insufficient in and of itself to produce a good applicant pool. If doing something that produces false negatives (which, yes, are undesirable) nevertheless convinces enough developers to consider your company who never would have considered it otherwise, then it seems like the advantages have outweighed the disadvantages.
Obviously, an even better scenario would be to produce the same level of developer interest without also producing any false negatives. Perhaps by investing sufficient resources in networking or in making his company famous to other developers, he could accomplish this. This would still be a possibility to explore in the future.
But just because his tactic has not produced the best possible outcome does not mean it has not produced a positive one.
The important point that I'm getting at, however, is that the dynamics of hiring shift dramatically as you move up the skill / connectedness scale. As you move towards the higher percentages potential employees are not competing for employers, but employers are competing for hires. And the dynamics of hiring are dramatically different at that point. If you're designing heuristics for hiring, you don't want to create a low-pass filter, and I believe that synthetic puzzles are exactly that.
The point of raganwald's post was on the surprisingly high cost of false-negatives. What I'm postulating is that your suggestion is just yet another way to reduce the false-positives rather than a way to reduce the false-negatives.