Well, if rbx became the default implementation, lots of the effort that was invested in MacRuby would go stale. There's an incentive for the MacRuby people to keep investing into MRI since they can more easily adopt improvements from there. The benefit they get from taking code from MRI will diminish over time, but at least ATM I'd still expect it to be significant.
(note: I'm not following MacRuby Development very closely at the moment.)
While MacRuby is a fork, they have already replaced the VM with one that compiles to LLVM, which has been implemented in C++. They have also gutted things like String and Hash to use Objective-C's (or more precisely, Cocoa's) counterparts, and they have started using libdispatch (aka Grand Central Dispatch), Apple's open-source multicore tech.
One of the big things about the new VM is that it eliminates the global interpreter lock, meaning MacRuby probably has much better parallelization than MRI. With the VM alone I think they are on their way to being their own Ruby implementation.
All of this is a bit disappointing considering that they could have shared their improvements upstream and improved MRI performance for everyone, not just for Mac apps.
(note: I'm not following MacRuby Development very closely at the moment.)