I am looking on my own statistics and more as 500.000 games my computers have played. For such a statistic its important to used many different engines.
Example:
1/52 if you used 20 different opponents / 650 games
1/31 if you used 18 different opponents / 650 games
1/24 if you used 16 different opponents / 650 games
One of the reasons I don't used clones.
Thats more or less an own statistic, not based on statistic programs like bayesian.
Sure, you have right if you used for statistics only programs like Bayesian. A great program!
Best
Frank
PS: With +-10 I mean 20 ELO different. With 1/52 I mean ...
In 1/52 cases you have a ELO different with more as 20 ELO, if you useing 20 opponents. In SWCR I have it with Naum 4.2. Ratings are stabil (20 ELO) after 610 games. Second record in my own statistics. Fist record for around 4 years with a early Glaurung version.
Optimal is to have 23 Engines in a list (each one vs. each one 50 games). You will get an ELO +-6 (12 ELO) after 920 games (if 40 games each one vs. each one).
Must not be right ... but after my own ...

Not easy to explain it in English for me. I should write a book about it I think, its a little hobby by myself.