Builds on the previous performance-related pull request with a more radical variation on the class version of memoize. This time it defines a method that replaces itself with an attr_reader before setting the instance variable using the supplied block. This makes the first call to the memorised method very expensive, but subsequent calls become very cheap.

In testing, this yielded an impressive performance improvement when a memoized method was called a very large number of times in a given instance's lifetime. However, the break-even point was about 200 calls, so it's not really practical, but I thought I'd put it up for interest's sake.