My thought is that it may be applicable if you're looking to find points of convergence, which classification of data by multiple classifiers may seek to do.

Would be a fun experiment. Do different classifiers who have been taught with different data sets tend to classify data points similarly? I'm unsure of an exciting real-world problem you could use where this would be beneficial over teaching one classifier with all of the data, but I'm also not super familiar with the math behind many of the classification algorithms out there.

Interesting. For example, with temporal data, like stock market data, some models might be "elders" in the sense that they are trained on longer spans of data, whereas, other models might be estimated just from the last few weeks.