Would have been interesting in the author of the original numberworld link had tried a different scheduler since that's what he blamed for the performance difference.
What are your thoughts on that link?
Well, the std::async thing is obvious(and there no mention of what version of gcc either), and since there is no information given on their homegrown "push pool" implementation, it's difficult to say. It's not open source, but from what they say and from their other source-available tools, I'd suspect that they're using C++ and using some high-level abstraction that gcc doesn't handle as efficiently as it could. That's a very common situation.
They're sort of jumping the gun by pointing at the Linux scheduler in my opinion. They should be looking very hard at MSVC versus GCC, especially if they are using newer C++ features...