That is a horrible comparison. It really depends upon how the code is written. Highly optimized C will have a tendency to be faster than GHC, but there are many notable exceptions. Compilers for functional languages have been really pushing ahead as far as automatic code optimization.
"Notable exceptions" are irrelevant. He is talking about basic performance differences generally observable between the two languages, not specific performance differences in specialised problem domains.
Haskell is a niche language, not a general purpose language. It is also an interpreted language, which are generally much slower than compiled languages. Again, exceptions need not apply. We're talking about general performance characteristics.
The benchmark I linked to is not authoritative. But I do not know of any other cross-language benchmark which uses: open-source compilers and interpreters, transparent testing methods and algorithms, well-known algorithms, and algorithms that are fully language-agnostic. For the purposes of basic, generalised performance comparisons between languages, it is a reasonably good benchmark. If a user wants to compare performance differences within a specific problem domain, they'll need to perform benchmark tests themselves.