It's interesting - I discovered the same thing a while ago.
Extension method invocations are described in section 7.6.5.2 of the C# spec, and it seems to me that all of the conditions there are met... because that talks in terms of an expression being implicitly convertible to the first parameter type (which it is, in this case). That suggests that either there's a bug in the compiler (possible, but unlikely) or that one of the earlier conditions for it being a valid method call isn't met.
I suspect the problem is that a lambda expression isn't a primary-expression in spec terminology - unfortunately, my spec-fu isn't up to much this morning due to a cold. A lot of the rest of the method invocation machinery probably relies on knowing the type of the target of the method, which isn't known in this case.
Note that the other conspicuous "untyped expression" has the same problem - you can't call extension methods on a null literal, either.
Very occasionally this limitation is a pain, but most of the time I don't have much of an issue with it - and I suspect it makes the compiler a lot simpler to implement...