Faster compilation speed

[Robert Dewar]

If garbage collection is taking a significant amount of time (is this really
the case), then concentrating on speeding it up may make sense, but I am
quite dubious that reference counting would speed things up (it very rarely
does, speaking of long experience in implementation of garbage collected
languages, because the distributed overhead is high -- one of the interesting
things about reference counting is that, since it distributes the overhead,
it then becomes very difficult to accurately measure the overhead.