Not arguing with you as that gets many good people during interviews. But I'm going to provide some anecdotal evidence to the contrary ...
In binary search this problem comes from the length of an array, which is either odd or even, and then you're doing a division by 2, which should trigger alarms in your head all over. And just as when you were taught in your first CS class (either high-school or college) you pick 2 simple examples for both cases, and test with pen and paper (tools provided during any interview).
When I did interviews from the employer's side -- the people not doing this were either totally unprepared, failing most other questions, or brilliant (and lazy) people that considered the problem too easy to bother checking the implementation.
Surely, put into context it can provide a valuable filter, but it's not saying much about the top X% of the talent pool, which my post was about.
The HN post led to this perennially popular article from Joel on Software:
http://googleresearch.blogspot.com/2006/06/extra-extra-read-...
Jeff Atwood references a good example: http://www.codinghorror.com/blog/2007/02/why-cant-programmer...
The sad truth is that many people who program professionally simply cannot so basic programming tasks. If anything, I think top 10% might be overestimating how many can do it - I suspect 10% might have some kind of idea of how binary search would work, but most of them will fail with edge cases etc.
I'm not particularly intelligent and I'm not in the top 10% -- but I could implement an in-place quick-sort in C in 10 minutes that my interviewer could run with only a minor fix (forgot a semi-colon), and I even described the parallel version with OpenMP (although there I got the syntax slightly wrong). This was for my interview at Adobe, when I got hired by them a couple of years ago; went on to other things since then.
By that logic that should place me in the top 1 percent ... but so are my ~ 100 friends that are also software developers from my city, and statistically speaking, something smells like shit in those statistics thrown around.
Maybe binary search, when discovered, used to be a hard to understand problem, but now it is taught from high-school. And sure there are lots of idiots out there, but many of those idiots also believe they are in the top 10%, because some statistic told them so.
So cut the crap and build stuff. Only by that metric you can prove yourself.
-- EDIT --
I'm not referring or addressing the article's author directly. I'm also not saying that you SHOULD be able to implement binary search, or quick-sort or whatever metric du-jour -- in interview conditions. I get it that you may be stressed by eyes watching you, or that you may be bitten by edge-cases other people haven't noticed for years.
I'm referring more to these metrics flowing around -- like, if you read HN you're in the top 5%, if you read this stupid blog you're in the top X%, if you can implement binary search ... etc, etc...
We are software developers, mathematicians, computer scientists -- surely we understand selection bias and should be able to recognize bullshit, even if it doesn't appeal to our ego.