I'm torn about removing the disable flag. This optimization requires the scheduler to do work creating the classes at the start of the cycle. If a site has all unique (or mostly unique) jobs, they'll end up with one class per job. This means the scheduler will do extra work for nothing. If a site knows this ahead of time, they can turn off this feature. While the scheduler could detect whether or not to use the optimization, it would take just as long to create the classes as to detect if the optimization is useful.
How much slower would things be in this situation? That would depend on the number of jobs. I don't really have an answer to this question though.
Now the other side of the coin. Would an admin of such a site actually use this knob? We have so many confusing knobs in the product already. Would an admin go through the trouble to learn what this knob does to actually use it? I'm thinking they wouldn't. It won't affect the end result of jobs being scheduled. The cycle might take longer, but would they notice?
Like I said, I'm torn.