I value code that's less confusing over code that allocates extra objects (largely because the last time I wrote some simple benchmarking code I could allocate multiple millions of an object in a second).
Now, I'm pretty sure that strings are interned in C#, so making them immutable makes a lot of sense from that perspective. It's just that the language encourages mutable state in so many other areas that it's a bit of a mental shift when you have to deal with strings. In Java... afaik strings aren't interned by default, so I can't really argue for immutability there.
The original author's argument for method chaining feels like a separate case altogether. Really it's primarily used when building embedded DSLs in these languages. In that case, standard semantics be damned; do what you need to to make the code look how you want.
otherwise:
means that name2 is now also "ANDREW DUCKER" when we quite possibly didn't want it to be.