Extremely poor cost/benefit ratio for what purpose? If your main goal is to adopt existing classes and libraries as your own (i.e. without any loss of functionality compared to language-native libraries), then extension methods obviously have an extremely high cost/benefit ratio. You can't analyze a feature in isolation of the language's goals. Features exist to serve a purpose, and are measured by how well they serve it.

To give an analogy, a shoulder-fired missile might have a terrible cost/benefit ratio if you're trying to hunt an antelope, but if you're trying to disable a tank, then I'd say it's cost/benefit ratio is pretty good.