I'm inclined not to use this approach, for these reasons:
1. As the author indicated, using the CLR within SQL Server opens up a number of potential performance and security problems. The author's code is quite admirable, but, I feel, ought to be executed outside the SQL Server environment.
2. DBAs generally are not facile at writing .Net code. Such would typically be a task for .Net developers, not DBAs. The author mentioned reviewing a T-SQL-based approach to JSON serialisation/deserialisation, but rejected it because of its length/complexity. Still, I wonder if revisiting a T-SQL solution might be worthwhile, in part for the reason of making it more accessible to DBAs?