Lazy evaluation probably adds more unintuitive cost. E.g. Haskell uses thunk to track yet-to-be-evaluated values. These thunks take up memory and take up CPU cost. For lazily evaluated function calls (or expression), each call has a thunk created and it will be evaluated later when needed. For deep recursive calls the thunk link can be very long.

Lazy evaluation is great in that the actual computation might not needed at the end but there are non-trivial cost to track and evaluate all the thunks.

It's an extra dimension to consider when comparing the cost of doing simple computation with strict evaluation vs utilizing lazy evaluation with the added thunking cost.