What makes you think I didn't? The author seems to agree with me. He even concludes that we should improve the implementation rather than the language for optimisation.
I didn't get that from the presentation at all. What I read is a guy complaining that idiomatic Python is necessarily slow because it depends, for its clarity and terseness, on not controlling allocations or memory layout, and that Python- the- language abets the problem by having its standard library build on the same idiom.
It's true that implementation makes a bigger difference to performance than language features, but language features can indeed affect speed. One clear example is that a language that checks for integer overflow will be slower than a language that doesn't. There are a lot of situations where there's no way you could optimize out the extra instructions you'd need to check for overflow.
Some language features are easy to optimize away for some common cases (e.g. array bounds checks, so that your program throws an exception instead of segfaulting), but you can't optimize these if, for example, you're iterating over data you read from a file using indices you also read from that file.
While I agree that implementations can be fast or slow, it does not completely eliminate the effect of the language. At minimum, different languages require different levels of effort to reach their optimum. So, while a language can be hurt by a poor implementation, it does not follow that all languages have the same potential performance.
As an example, just consider the enormous amount of effort that has gone into the JVM, and Java is still generally considered to be ~2x slower that C.
Didn't it occur to you that these slides were for a presentation and that sharing them enable more people than just those that were at the presentation be informed of their content?
I, for one, am very grateful to speakers that make the extra effort required to share with a larger group what they have already shared (or are about to share) with a smaller group.