I took a stab at separating out the method dispatch costs from the cost of the #extend call here https://gist.github.com/4436640 https://gist.github.com/4436640

I took a stab at separating out the method dispatch costs from the cost of the #extend call here https://gist.github.com/4436640 https://gist.github.com/4436640. I also tried making a call against an unmodified instance of a completely different class to demonstrate that the global method cache is indeed flushed in MRI by doing a runtime #extend on any object instance. Subtracting the results of different runs like this is pretty questionable, though (the errors accumulate), so the above is to be taken with a pinch of salt.