Good post, thanks. One comment:

It may appear that the partitioned data always give a better answer than the segregated data. Unfortunately, this just isn't true.

First, I assume you mean "aggregated", otherwise this statement doesn't make sense.

Second, I don't believe you. I say it's always smarter to use the partitioned data than the aggregate data. If you have a data set that includes the gender of the subject, you're always better off building two models (one for each gender) instead of one big model. Why throw away information?

There is a nugget of truth to your claim, which is that sometimes the partitioning strategy becomes impractical. To see why, consider what happens when you first partition on gender, then on history of heart disease. The number of partitions jumps from two to four, meaning there are fewer data samples in each partition. When you add a couple more variables, you will have more partitions than data samples, meaning that most partitions will be empty.

So you don't always want to do as much partitioning as you plausibly could. Instead, you want to figure out how to combine single partition statistics corresponding to each condition (gender, history,etc) into one large predictive model. This can be attacked with techniques like AdaBoost or MaxEnt.