My conclusion: Although this patch is useful for me and doesn't cause any immediate problems, it is nevertheless unsound if left as-is, in particular because ClassNode.methods won't reflect changes to method names. Therefore I suggest to revert the patch and apologize for the work this has caused. In retrospect I think the patch has fulfilled its purpose by starting a good discussion (thanks to Paul).
I will propose an improvement for 1.7 to support arbitrary method names, or at least all method names allowed by the JVM spec. Especially the latter should be fairly easy to achieve - it already seems to work now unless the method name becomes part of a class name, for example due to the definition of a closure. Another improvement would be to disallow unsupported method names in the first place in order to avoid strange errors later on.