We'd like to. The difficulty is that in order to get the performance improvements we had to integrate three different components: a few patches to libcoroutine, a patch to glibc, and pooling code that's tightly integrated with our asynchronous IO layer. A compelling patch would span three different licenses, involve patches to two libraries and re-engineering of the integrated piece to make it usable by the outside world. We'd like to do it, but for now this would be too much of a distraction.