Emdros uses, internally, a data structure known as a "Skip List". It is basically a glorified linked list, with randomization applied to make it look somewhat like a balanced binary tree. It's efficient both space-wise and time-wise for a range of problems, though it often cannot beat a really good implementation of a red-black tree, for example.

Skip lists were invented by William Pugh, and described in a paper from the early 1990'ies. In the paper, Pugh described the various options for the randomization to be applied to the data structure. Based on Pugh's recommendations, I originally chose a particular kind of randomization, but failed to experiment with different kinds. Until today, that is.

I found that by simply tweaking the number of bits to consume from the random number, as well as raising the number of elements catered for in the data structure, Emdros as a whole could be made to run consistently about 5-6% faster across my various test suites.

That's a lot of speed increase in exchange for three lines of different code.

Incidentally, while running the tests, I found that the BPT engine (my proprietary backend engine) is still at least 30% faster than the SQLite 3 backend for the same database content and the same set of queries.