Any system can get messy when the project size grows. Defect rates seem to be correlated with the addition of new features. Stable projects only including bug fixes with good test coverage become more stable over time. This is all regardless of language.

There may be a case that typed languages ship with fewer initial defects in new features but I'm not entirely convinced it's always competitive with the return in development speed and simplified testing you can get staying untyped. Particularly when your underlying problem is not CPU bound.