Why update at all? There are still legacy systems using FORTRAN and probably COBOL as well. While there are C#, Java, PHP developers all over the place I imagine that finding a developer to maintain a legacy system is extremely hard. Of course that means there will not be many jobs out there for you but the pool of qualified applicants will be extremely small.
Plenty of money in COBOL but there is a need to suit up (physically and mentally) - not for everyone.
I have found a small but significant niche in embedded *nix programming. Small yet powerful systems requiring every scintilla of juice tempered with a familiar API - C Systems programming work is common enough (yet not common enough!). This is where I hope to spend the next while.
An "old school" approach to knowing the architecture inside-out and attention to detail is clouded by the bizarre abstractions of C# and Java. PHP isn't even an abstraction, it's a distraction (I grew tired of the inconsistency so no longer practise).
Perl is unfashionable in some circles and has a reputation for having magic constants (or whatever it is the detractors call "I don't want to learn this language") but I recommend it if you want dynamically typed "chops".
I find these "chops" are overrated. I enjoy low-level thinking so don't need to bloat up with virtual machines[1] (the real ones work fine for me), OO[2] (I know how to pass a pointer to my data to a lib) or design patterns[3] ("ways to do things" - if you learn one way as "the way" you may be unlikely to think there may be a better way)
[1] I use virtual machines but it's perverse running the dozens of MB JVM (and waiting around for it) for a browser bound animation or trivial desktop app. There may be a better case for this messing on the application server, but I don't care.
[2] OK, I will make an argument for OO in GUI programming - a large and complex library of heterogeneous components is difficult to arrange sensibly in a procedural manner. gtk_status_icon_set_from_file(foo_icon, "bar.png") or fooIcon.fileSet("bar.png")? There may be a similar argument to be made for other systems but for the most part I find the OO model a needless abstraction.
[3] Right... most programmers aren't brilliant - I know I'm fucking terrible for the most part - so having established methods for common situations is no bad thing... just don't get too attached.