[Fernando Tella]

I have a scene with some grass made by instancing grass blades (well, yes, I tryed that). It produces very low benchmarks, but that's not the bug I'm talking about; while making a region render of an image I found out that hiding the layer containing the grass instances (expecting higher benchmarks) doesn't actually speed up the rendering unless I uncheck the "Use instancing" checkbox. So I guess that it still makes the calculations due to the instances even when they are not rendered and their layer hidden.

[Bogdan Coroi]

Could you share the scene so we can check hidden layers for wrong exporting time?
Care to detail your low plug-in benchmarks?

[Fernando Tella]

I'll try to prepare the smallest possible scene for you. Or maybe you prefer the whole thing (138MB)?

Also found out that it's faster (at least in this scene) to render with "Use instancing" unchecked, if you've got enough memory, than doing it with it. The benchmark is the same at the beginning of the render but increases when goes through the SLs (but this would be a mxcl problem, I guess).

The benchmark with the Quad (check my sign for details) changes from ~5.x with the "Use instancing" checked and instances hidden to ~25.x without it checked.

If I render without hiding anything and without using instances the benchmark starts at around 5 and goes up till around 25 (which I consider ok due to the high bunch of clipped planes and ivy).

[Bubbaloo]

I have found that intersecting clip mapped planes and instancing really slows things down.

[Bogdan Coroi]

If you could simplify the scene it would be perfect.

Regarding your benchmark:
Yes, your guess is right. The plug-in's job is to export geometry and the only benchmark relevant for the plug-in(s) is the time it took to export the scene, with or without different settings enabled, like "Use instancing" in this case.
Since MXCL is rendering the exported scene, all rendering benchmarks are MXCL related.

[Fernando Tella]

After some test I think everything is working ok except that my instances were not editable meshes but scatter objects + uvw mapping. That alone makes that a render with instances produces lower benchmarks than the same render without using instances.

Sorry, I got confused by this. Any way I'll make some more test to get completely sure.