Different languages are good for different things.

At SLAC we use Matlab extensively for accelerator modeling and control, It has a lot of very nice features (with add-ins) for signal processing, feedback design, linera algegra and general numerical analysis (numerical integration etc.). It has some nice parallel processing tool boxes and can be very high performance for vector applications. I also has excellent graphical and debugging capabilities.

I've used Matlab and Python (Pylab, numpy) and find Matlab to be a much better tool for this type of signal processing problem, and worse at other jobs. I haven't used R, but I believe it is optimized for statistical analysis which is a different sort of problem.

The wide variety of tool boxes are a big help for many types of problems.

Like most specialized languages it is very good at doing what it is designed to do, not so good for other applications.

The cost of matlab is not significant compared to the cost of the person who is using it. (A FTE here costs the lab ~$250K/year, so a few thousand for a Matlab license is lost in the noise).