We implemented a version of this algorithm in microcode that runs inside of a GE in an Onyx Reality Engine. We created subdivision surface versions of a NURBS teapot and a car part. Even though our microcode was in C, not in assembler as most of the OpenGL, we were able to beat the OpenGL NURBS in rendering speed. The car part was actually a trimmed NURBS surface: we could obtain similar effects directly by introducing holes to the subdivision surface.
What was a bit unexpected was that our algorithm beat NURBS also on an Indigo 2 where it runs on the host, rather than inside of a GE. This shows that this is an efficient algorithm also on uniprocessor machines.