Lists in Haskell are special in syntax, but not fundamentally.
Fundamentally, Haskell list is defined like this:
data [] a = [] | (:) a ([] a)
Just another data type with two constructors, nothing to see here, move along.
The above is kind of a pseudocode though, because you can't actually define something like that yourself: neither [] nor (:) is a valid constructor name. A special exception is made for the built-in lists.
But you could define the equivalent, something like:
data MyList a = Nil | Cons a (MyList a)
And this would work exactly the same with regards to memory management, laziness, and so on, but it won't have the nice square-bracket syntax (at least in Haskell 2010; in modern GHC you can get the special syntax for your own types too, thanks to overloaded lists).
As far as laziness goes, that's not special to lists, but is special to data constructors, or more precisely, pattern matching on data constructors. In Haskell, every computation is lazy. This means that whatever crazy chain of function calls you may construct, it's not evaluated right away. Doesn't matter if it's a list construction or some other function call. Nothing is evaluated right away.
But when is it evaluated then? The answer is in the spec: a value is evaluated when somebody tries to pattern match on it, and at that moment it's evaluated up to the data constructor being matched. So, for lists, this would be when you go case myList of { [] -> "foo"; x:xs -> "bar" } - that's when the call chain is evaluated up to the first data constructor, which is necessary in order to decide whether that constructor is [] or (:), which is necessary for evaluating the case expression.
Index access is also not special, it works on the exact same principle: the implementation of the (!!) operator (check out the source code) repeatedly (recursively) matches on the list until it discovers N (:) constructors in a row, at which point it stops matching and returns whatever was on the left of the last (:) constructor.
In the "simplest" lambda calculus, in the absence of data constructors or primitive types, I reckon your only choice is to Church-encode lists (e.g. as a fold, or directly as a catamorphism) or build them up from other structures (e.g. pairs), which are themselves Church-encoded. Lambda calculus, after all, only has functions and nothing else. Unless you mean something more specific.
