BLAS and LAPACK are libraries.
You need to differentiate between the dynamics of a language (ie: FORTRAN vs. C/C++) and the libraries available.
FORTRAN 77 vs. FORTRAN 90/95 and up are completely different species.
So we start talking about eigen systems programming in one language vs. another. Well, when was that library written? In what version of what language? Just because it is a widely available library, does that mean it is any good internally?
Theoretically, if there were a fully C++ written linear algebra (or any other library) that isn't linked with some gawd aweful old FORTRAN code or (asm{ ... }) down in the bowels of the machine, then you could make an honest comparison. But since everyone seems to start off with poor examples from free programming cook books and someones opinion from the web, without seriously (re)designing or understanding the patterns used to accomplish the task, you then get what you get (ie: crap).
After long time programming in FORTRAN 77/90/95/etc. and C, and C++, and many other languages I would have to say that most programming comes down to energy expenditure. If a grad student comes out of school after programming mainly in Matlab the first thing that person is going to suggest for a programming project is going to be Matlab. This same phemomenon is what has kept FORTRAN alive. In the case of FORTRAN , the legacy dependency code of many scientific applications ultimately led to the refactoring of FORTRAN as a language rather than discarding all that code. It amounts to loss aversion and an unwillingness to learn new languages in entrenched users.
Why not create an open scientific co-processor card spec that has hardware advanced functions instead of farting around with GPU discretes that were originally designed for video games. Then we could just have linear algebra calls in the standard math library that are driven by math hardware instead of 50 years of accumulated CPU work-arounds for 8088 code (that was sarcasm).
Progamming always seems be 'VHS instead of BETA' because most programmers doing applied programming for science arrive in industry with only single language skills and programming was only a sideline from whatever thier degree was in.
I also continue run into 'C' programmers who refuse to learn C++ . it's some kind of religion thing. Deities will apparently smite them if they crack a manual.