Some good news with the export speed I think
Here are some results...
Test with a 1.000.000 triangle sphere.
Export from Softimage.
Old:
untextured ~35 sec.
textured ~40 sec.
New:
untextured 16.6 sec.
textured 22.8 sec.
Import in Octane:
untextured 22sec + 15sec voxelizing.
textured 33sec + 17sec voxelizing.
Test with 50.000 cubes.
Old export from Softimage, ~79sec.
New export from Softimage, 22sec.
Import in Octane, 20sec + 5sec voxelizing.
Have found that the most time is spent in the conversation from values to strings.
Without the conversation, i can export in the half time, without any values

Have also found, that many small objects needs more time as one big object.
Also a multithreaded version make no sense on many small objects, maybe on a big one...
In the link, that are postet from jamnique, Silo uses << endl to begin a new line. The programmer uses \n to make the same and save with that the time.
I don´t use it, so the time i saved with stdio to iostreams is nearby 0...
He has an exporttime from 1.3-1.4 sec on Silo with a 50k poly mesh, i don´t know if quads or tris.
I have 0.8 - 0.9 with 100k tris
For PointClouds, I write for every point a new mesh. if I make one mesh, it should be faster, will test it now
face