Sather (and Eiffel and Smalltalk etc) are less broken than Simula, I'll grant you that.
I do come at this from the point of view both as an engineer working in the high-performance computing area, and as a programming language theorist. People like Stepanov, Knuth, Dijkstra, Luca Cardelli, Alan Kay, Benjamin Pierce, and many others have written and spoken (sometimes at length) on the problems with implementation inheritance. And the fact that most OO proponents consistently tell you to avoid it ("prefer composition over inheritance") is telling.
However, I would say that the claim of "broken concept" is based not on one single argument, but on independent converging lines of evidence.
Here's another one, which again is not convincing by itself, but adds to the picture: the Curry-Howard isomorphism. One of the signs that you know you've found something interesting is that it turns out to be formally equivalent to something else that's interesting. (Think of the isomorphism between regular languages and DFAs, for example.)
Programming language theorists have found many such isomorphisms over the last few decades, and what's interesting is that programming language features seem to be equivalent to interesting objects in logic and category theory. For example, Scheme-style continuations are actually Pierce's Law in logic. Call-by-value and call-by-name turn out to be dual in the programming language which is isomorphic to Gentzen's classical sequent calculus.
Subtype relationships are pretty well-understood, and interface inheritance has a straightforward interpretation (e.g. see Haskell's typeclasses). However, despite searching for decades, nobody has found any such connections with implementation inheritance.
The industry desperately needs the style of OO that everyone uses to be on a sound theoretical footing, because it makes program analysis and compilers better. Lots of smart people have tried. The fact that we haven't found it by now strongly suggests that it's not theoretically sound.