I have documented what I think is a bug. Someone test with a new version of MRSA please.
Three assemblies (torus, cone, sphere) with X translate set to -25cm,0cm,25cm respectively.
Unified sampling returns expected result.
Bounding Box Update set to “None”, however all modes of Bounding Box Update made no difference.
Also, rotation of assembly off orgin returns an unexpected result.
Mental Ray Standalone ver. 22.214.171.124, Maya 2012, MRSA Python Script ver. .9
It appears it's taking whatever value for the offset and doubling it. Rotation is another story.