Personally, while I do find DSLs easier to read, I find writing them harder than recursive descent. The problem is that a DSL will still have theoretical limit

Personally, while I do find DSLs easier to read, I find writing them harder than recursive descent. The problem is that a DSL will still have theoretical limitations (ie shift shift or shift reduce errors, left recursion, performance). In order to deal with these limitations you still have to know how everything works under the hood. With recursive descent, you can always start adding break points until you understand the forces involved.