Additionally, it should be noted that mainframe IBM COBOL has also continued to be updated and extended, to the extent that it has native support for JSON serialization/deserialization for building web services.

There is an enormous amount of new COBOL being written even for mainframes.

There is almost certainly a large market for COBOL on .NET if it managed to replicate some of the functionality IBM added on top of the standard.

A better term for COBOL (and mainframes for that matter) would be niche, not legacy. It's the best language for a very particular subset of business problems, and there hasn't really been much attempt to replicate that functionality by newer languages, and so COBOL remains in use. There simply is not much overlap between the folks coming out of school interested in language design/theory and those that are aware of COBOLs continued dominance in certain types of business.

It's also ruthlessly efficient compared to something like Java or .NET, and even beats out C by a wide margin in certain scenarios. That isn't as important if your not laying mainframe licensing fees, but it is a common headache that makes migration to some of the current virtualization based modernization platforms a headache.