John Dunning has produced a fantastic blog entry summarising his perforamance characteristics research at EBRD, the European Bank of Reconstruction and Development - "Benchmarking Drools" which was a follow up on his "Banking on Drools" blog. In this he takes a simple problem and solves it a number of different ways, benchmarking each, he details all the variations with several perf charts comparing the approaches. Someone else has recently taking a variation (making it harder) of his benchmarks and run it under JRules, with very surprisingly results, I'll provide more details later on when the results have been better verified :) I'll also try and get a zip where you can run this benchmark for Drools and JRules yourself, as luckily ILog have now made their software easily obtainable for trial purposes. Lets's put it this way, which system do you think scales easily to over 500K, even 1mill, objects, and which one didn't :)
Steve Shabino has started a blog on his experience with Drools, as they have an extreme problem to solve involving the need to reason over 2mill objects in a stateful session. I'm really looking forward to seeing his findings.
The mysterious Yaakov Kohen has started blogging again with an insightful blog on the problems with todays acadmic benchmarks.