Excellent scene for bi-directional integrator to see difference. That yellow cover cut Cycles direct light probes (work only with Transparent BSDF), and you get noise from indirect light because it can hit bulb very rare. I make quick test to prove that, delete all point lights, replace all complex shaders but basic diffision/emission and translucent for cover, noise on wall in case of bi-directional clean very fast (it expected, light paths starting from bulb are very dense in space near lamp), even after 10 samples. But noise on cover still noticeable and very different from uni-directional, maybe i have bug, much check.
That strange difference with Transparent BSDF hold me from call for wider testing of bi-dir patch, i hope one-two weeks later will found bug and start flooding threads with special twieaked cheated scenes that show bi-directiuonal advantage
Scane by MadMinistrel, all point lights deleted, all materials are basic, yellow cover is pure translucent, light bulb emission bsdf.
Warning, it is not fair to compare uni- and bi- directional with same max bounce settings, bi-directional use more bounces, in my half-finished case it 1.5 more with same settings (i add only pure light->camera chain), when if will feature complete it will be more like N*N bounces, and you better compare time to clear noise then quality of pictires with same sample number.