In my experience, the reasons are mostly just because it's been the standard for so long at said shop.
Most people using mainframes don't make money off of their IT department, their programmers are expensive and don't produce profits. Investing in a large change from "what works and has a fixed cost" to something new and possibly troublesome or more expensive (initial cost) just isn't a priority for them.
Not to mention the retraining you'd have to do. A lot of these programmers are either old and unwilling to learn, or just uninterested in rocking the boat. So is it worth it to have to fire people or get subpar code, or just stick with what's working that keeps you generating a profit and keeps your current programmers job security?