edit: um...what? why is this getting voted down ?
I'm afraid it's because of my fat fingers:
-was on the way home
-watched this interesting thread on my phone
-wanted to vote up
-but somehow it was detected as down vote
-I cannot correct that anymore. I'm sorry.
But your results are very interesting, it's rare to see here some actually profiling numbers. yet, your test case is not very tile based rendering friendly. It's just one object at a specific place, if you slightly zoom out, it could actually fit into a tile. And that's actually what your results report, you are about 10ms slower than without the optimization and your binning takes 11.4ms.
A better test case would be if you'd render 32x32x32 boxes, covering most of the screen, then you'd actually become framebuffer bound and your tile based framebuffer optimization might kick in.