And even the new explanation, I STILL don't like MVR. I think that returning a list or other data structure, or just writing two functions, is much clearer, and lends itself better to function chaining. Sure, in CL there's a default, which makes things a little better, but what if you want the other one? I don't want to write
(multiple-value-bind (a b) (foo bar baz) b)
Generally speaking, taking a secondary value alone feels like a bad use of MVR (it happens rarely with standard functions and most libraries).
MVR are not intended to be substitute to other data-structures. As said elsewhere, you can and should use structs, classes, lists, hash-tables,...
After all, why do we use multiple values instead of a single struct for functions arguments? Because in CL and other languages, unlike in Haskell, you don't want to define an ADT for all your functions or rely on currying. But if you start putting too many coupled parameters in your functions, it starts to smell fishy.
Extracting a single value is easy, by the way:
(nth-value 1 (foo bar baz)) CL-USER 76 > (nth-value 1 (values 'a 'b 'c 'd))
B
(multiple-value-bind-some (NIL b)
(foo bar baz)
b)
This is actually one of the cases where syntax-rules would allow you to write something cleaner, I would guess. I like syntax rules, but everybody always seems to complain about it...