I have multi-CPU linux server, so I wonder what is the fastest version of gdal2tiles.py ?

My Ubuntu version:

lsb_release -a
Description:    Ubuntu 12.04.5 LTS

How I install the package:

sudo apt-get install python-gdal

Package version:

apt-show-versions python-gdal
python-gdal/precise uptodate 1.10.0-1~precise2

And under /usr/bin/ I have gdal2tiles.py.

Does it make sense is to update python-gdal package > 1.10.0(maybe build newer version from source), where I can see changelog about gdal2tiles.py?

After some googling I found that there are some patched versions of original gdal2tiles.py and some parallel versions of gdal2tiles.py. So I wonder why this functionality is not included in official package?

If anyone point to stable and fast version of gdal2tiles.py I can run some benchmarks for community.

Here is my benchmark for some dataset using original single threaded gdal2tiles.py.

set -x # Dump each line of this script
set -e # Stop on first error

DATASET="my_dataset"
LEVELS_OF_DETAIL=19

time gdalbuildvrt -srcnodata "0,0,0" orthomaps/$DATASET/global.vrt orthomaps/$DATASET/orthophoto_*.tif

#Default gdal2tiles.py
GDAL_TO_TILES="/usr/bin/python /usr/bin/gdal2tiles.py"
time $GDAL_TO_TILES -z 0-$LEVELS_OF_DETAIL orthomaps/$DATASET/global.vrt default_gdal2tiles_$DATASET

What versions I have tried:

1. Original version of gdal2tiles.py (single threaded).
2. Patch from here to make original gdal2tiles.py parallel, produce missing tiles.
3. Version from here (also parallel), produce missing tiles.
4. Version from here (also parallel), I can't even run it.
5. Maptiler, works well, but cost money.

Related question: Performance of google map tile creation processes