The current refill path for tlabs calls all the way into the runtime when it should be simple to write a stub to do the work as if it was a normal fast path allocation. Preliminary testing indicates that it removes many of the performance downsides of using tlabs, at least for c1.