Optimize
The benchmarks have been run by running bst show on the debian project (no target listed), and recording only the 'loading elements' time.
bst show