As I understand it, scoping is handled by "contexts", which can be inconvenient if you expect every function to be a single standalone snowflake, but just fine if you`re tossing off quick scripts which don`t require hermetic degrees of encapsulation. You can also force scope by using 'let' or declaring private variables as arguments, among other things. Certainly Emacs Lisp seems to do OK with similar constraints.
As mentioned elsewhere in this post, the deep copy thing may not be as bad as you make it out to be -- certainly it's never particularly troubled me.
As for cons... It seems to me that cons is a rather obsolete construct, referring to a long obsolesced hardware feature. I can't get too unhappy about someone trying to make it more immediately useful in a modern context.
We've talked about what you don't like about it -- here's what I think NewLisp has gotten very much right:
* It's simple to use.
* It's small and self contained.
* It has a REPL and you can do introspection.
* It has conveniently and reasonably named built-in functions.
* It deals well with a UNIX-derived file system.
* It's easy to pick up and start using, but you can still do fancy stuff.
* The documentation contains worked out examples for every built in function.
* The main developer is very responsive and helpful.
* Trivially cross-platform with downloads, not compiles.
I have found NewLisp to be very convenient for many of the problems I have applied it to. I have used it for everything from dynamic web pages to image processing to ad-hoc render farms. So, I certainly feel that it may be invaluable for someone who is interested in Lisp, but has been intimidated or discouraged by the "official" compilers and needs to get some real world work done.