I tend to avoid stack based recursion when the application/routine could require unbound traversal depth (eg: depth is controlled by some non deterministic input, like user data). The last example that comes to mind was building an new enterprise financial rebalancer which needed to support dynamic asset model resolution. In a model, assets were dynamically associated via tagged characteristics. Models also supported including other models as children, so they were well represented with a stack like structure. This was a .NET based code base, which meant StackOverflow exceptions couldn't be caught (they immediately terminated the process unless some icky work arounds were used). For those reasons, we used heap based recursion to walk and resolve the final in memory model structure.