Breaking the PLT Scheme optimizer by mutating pairs it thinks are going to be immutable and then going in and figuring out which optimizations to fix doesn't sound very simple to me... but, assuming that it was easy to do, what does starting from 4.2 do for us?
I guess I just see "Take the newest MzScheme and implement not-so-immutable conses again" as a conceptually easier task than "Take the fork and implement feature A and feature B as inspired by the newest MzScheme." Of course, the more objections we have to MzScheme changes, and the more features of our own we want to include, the harder this becomes.