To be honest I don't think FarCry is currently a good engine to use for benchmarking. The behind the scenes tweaks the engine performs is currently too opaque to have confidence that we know what the engine is doing.
Consequently the variance we are seeing can simply be a result of the mixtures of rendering paths the game silently chooses when initialising itself. We also have to factor in CPU loading as the detail settings within the game also add extra animation and AI load.
It is CPU dependant a lot though. Also, I think I read the 1.2 patch will fix the problems with demos not replaying gunfire back. That will help.. because as we know, gunfire and the effects it has can drop frames a lot.
