On top of that they tend to be stodgy old corporate companies that developers hate, where you have to wear a suit to work and being seated by 9AM is more important than doing your job well. Then there's the fact that it could be career suicide to spend so long on tech that fewer and fewer companies are using. And you'll need an exit plan because it's used at become companies that do regular mass layoffs, they've probably fired more cobol programmers than they'll ever need. They should be paying a hefty premium to get people.
The language itself and surrounding systems can't be any worse than the standard enterprise XML DSL rubbish can it?
That I can attest happens, less so today but then such places lost most of their talent as those that could and able to move on did and those that didn't, well they would of been close to retiring and with that. Hence technical debt increases. But it's not just knowing the COBOL, it's not just knowing that business logic, but how the two go together - THAT level of knowledge was always undervalued in many institutions alas.
MARK IV is the oldest language in use that I know of that I have personally had contact with, and I've had contact with it recently. I would say that the issue isn't just finding Cobol programmers, but a mix of mainframe specific languages and technologies no longer in wide use outside of mainframes' data warehouse solutions, such as DB2 and VSAM.
The main problem with Cobol, REXX, JCL and MARK IV is that the solutions/problems encountered cannot just be google'd through stackoverflow or some online forum. Institutional knowledge from 30 year experienced self-starters or solution specific candidates with 30 years of experience in a piece of the project needing assistance who is willing to be a self-starter for gaining experience enough in the other technologies the mainframe uses.