Minetest showed fine results with about 60 fps... what is the reason for slow fps in irrlicht and how can i improve it? I'm using a fairly standard setup using an IMeshSceneNode w/ an SMesh consisting of an IMeshBuffer with all the vertices and normals that the surface extractor uses.

This performance problem is unlikely to be related to PolyVox so your probably better off asking in the Irrlicht forums. Are you sure you're not accidentally running the extractor every frame, or uploading the data to the GPU every frame? And you're runing a release rather than debug build?

I'm probably uploading the buffer data to the GPU every frame... I'll look into this more, I'm not too experienced. Thanks for helping me, I know this isn't the Irrlicht forums.

EDIT:
My program was set to debug build. I'm rerunning it with a release build, but would this really slow it down a lot? why?

Yes, this can have a significant effect. When in Debug mode a lot af additional checks are performed, such as validating memory accesses to make sure you don't go outside array bouds. Also assert()s are typically active (PolyVox has many of these) and optimizations are turned off.

It's possible that's you'll see a frame rate difference of 5-10x, though this isn't guarenteed.

Just to contribute a bit here: In my RawVolume-per-chunk with data overlapping engine (with all of its complications) + some libnoise stuff... in Debug, 16x16x128 chunk "loads" in 4-5 seconds, and in release - 2-3 chunks load in a second. Using Ogre, frame rate with 49 chunks looked from above is: Debig: ~200 FPS, Release: ~250 fps. Looking straignt up makes 1000 fps in release and ~500 fps in Debug (1000 is the FPS cap).

TODO: compare ANL vs LibNoise.