One thing I often bring up is if you do scientific computing you probably use code written in Fortran as Lapack is written in Fortran. One thing that is underappreciated is that Fortran compilers are really good at optimizing array computation, and lots of numerical code is scarcely more than just that.
Because APL is an opaque nightmare for a newbie, despite having a better matrix handling story?
Plus the huge range of pre-existing mathematical libraries of course.
Putting aside the inherent user-surliness of APL –
Fortran has some really fast compilers available for modern hardware. APL lags in this regard.
Because of ‘switching costs’: there’s a ton of code and a lot of knowledge out there that would be expensive to replace.
I can’t recommend “Information Rules: A Strategic Guide to the Network Economy” ( http://amzn.to/2ys28sO ) enough. As they say in the introduction “Technology changes, economic laws do not.”