Yes, just like having a different windows and posix implementation; the architecture for multiple I/O backends exists.
As stated, the rate of security issues discovered was in earlier versions of the code. Also: The 30% less CPU cycles were of an entire (GSM Base Station Controller) application. If you’d run a low-level I/O benchmark that does nothing but I/O, the speed-up would of course be much higher than that.