Old platforms never die, they just go into maintenance. Even mumps is still around in quite a few hospitals.
MUMPS (har har) has it's origins in the medical world so there are a lot more people versed in mumps in that world that have domain expertise regarding hospitals but you'd hope that that chapter of the Cambrian explosion in programming was closed by now and that someone would come to their senses and would say: absolutely no more new development in MUMPS. (And 'Mapper' for that matter.)
They were dead-ends 30 years ago, they're not magically going to be better in 2015. Of course it would be hard to get rid of 1.8B euros using any other available technology so maybe that's were we can find part of the answer of why this happened.
Maintaining MUMPS code is an interesting exercise in pasta consumption, the language more or less dictates it.
See also: http://thedailywtf.com/articles/A_Case_of_the_MUMPS
Anyone want to build an LLVM backend + FFI module? Probably worth a small fortune to shops that want to migrate out of MUMPS
But I agree with you that if you can pull this off you'll be doing well financially.
Even better: a MUMPS -> Java automatic translator (or maybe Python or Ruby, but I think the Java would be an easier sell in that market).
[edit: specified which summer]
Seriously, dont.
One of my worst experiences was working for some bank that used MUMPS and trying to recover the (accidentally) erased editor string before someone would notice that it was gone. (Yes, the editor is a string, and it behaves just like any other string and globals are auto persisted to disk so if you wipe the editor string then much good luck to you.)
It's a disease primarily limited to the healthcare system, in my experience.