Is possible to improve likelihood of superiority resolution?

[cdani]

I'm sure someone has thought about this before, and as nobody is using it, it didn't work. But anyway I ask if one can try to improve (get it faster) the search for a better engine version adding to the number of won/lost games some ponderation of the number of moves where the two engines thought that one of them had a better/worst position evaluation.

[Fabio Gobbato]

I tried in the past to collect statistics of the searches for tuning.
For every search launched by the engine I stored in a file the depth reached and the node count.
The idea was that the better version should go deeper and search less nodes.
But it didn't work.