My thought was that most binary distributions probably use very conservative configuration that will generate code that compatible with very old processors, and that you would therefore not see much use of modern instructions in /bin. This is one of the selling points of compile-yourself distributions like Arch/Gentoo: you know what processor you're running on, so you can take full advantage of its features.

That doesn't really take away from the finding though. This is obviously not meant to be a thorough analysis of what opcodes a compiler "knows" (he could have just looked at the source code for gnu cc if that were the case), but rather a quick and dirty hack showing the opcodes in actual use. Furthermore, I would assume that handwritten or inlined asm is an insignificant portion of all the opcodes in bin, making it statistically irrelevant for the pie chart. Also, handwritten asm could only add opcodes to the count of unique opcodes in bin, meaning those 200 are not only not used by the compiler, but also not used in handwritten code.