Multi Scale Volumetric Obscurance much slower than Scalable Ambient Obscurance

Despite what the documentation says, I'm finding MSVO much much slower than SAO on a modern desktop (GTX 1070, updated drivers), at least in the Unity editor. The quality/intensity settings don't really matter that much. MSVO is about 50 times slower (1.92ms per frame for MSVO as opposed to 0.04ms for SAO for the scene I'm looking at now).

Is this expected behavior?

I'm using the version of Post Processing V2 on github.

 

I'm seeing 3.5ms on PS4 and 8ms on Switch. And the documentation recommends it for mobile?

 

Does the documentation actually recommend it for mobile? If so, please ignore that No idea what Unity would be thinking to add that.

Multi-Scale Volumetric Obscurance, put simply, is multiple passes of AO across a range of radii. Scalable ambient obscurance is a very cheap method that requires a bit more tweaking of values on a scene-by-scene basis but can produce great results in many scenes. For mobile, I would only touch SAO, or optimized versions of NNAO (Neural Network AO, like that found in my post-processing asset, PRISM).