There have been many suggestions as to the problems for the vendors at October Rules Fest this year. I have submitted one already (Einstein's Puzzle) and the "beginning" of another (Miss Manners 2009) but I have not done the data on the last one. I really was hoping that an enterprise-size vendor (FICO, ILOG, Haley, Pega, ???) would find someone on staff that is not too busy and task them with this simple but time consuming job. Didn't happen.
Geoffrey De Smet suggested that we use the benchmark problem from the ITC competition, http://www.cs.qub.ac.uk/itc2007/index.htm that had to do with Exam Registration. On the benchmarks page they said that it was running between 300 to 500 seconds on a "modern PC using Windows XP." I would think that 5 - 10 minutes would be a bit long for my tastes but you never can tell. Personally I prefer old-fashioned Unix (or UNIX if you're that old) rather than Windoze or even Linux.
I have looked at this particular problem / benchmark and it appears to be more of an Operational Research (OR) problem in optimization than a rulebase problem. We "could" require that the solution be done in a rulebase but those who have OR engines would scream bloody murder if we did.
However, I leave it with you in the interim to determine what should be a good rulebase problem and what does not lend itself to simple "data smashing" of a typical "business" problem. The problem should be fairly complex and require a bit of work on the part of the engine itself rather than depending on the operating system and database to do the work. Please either comment (or if you have authorship privileges then blog) some kind of response to this. Whatever is decided should be a problem that is long on complexity and short on blandness.