Avi Bryant writes:
 > 
 > You could certainly use the optimization until the first time #class is 
 > redefined, and then recompile all the methods that send it...

To me, optimising #class sends look similar to new sends and at: sends.

A better optimisation might be polymorphic inline caches with
inlining. That would also be a nice way of speeding up normal sends as
well. Of course, these optimisations are best done by a compiler...

The type specialised code should be fast. Each call site is only
likely to get a few different receivers.

That would both speed up the currently optimised case further and
speed up other cases (normal message sends) as well.

Bryce