Relative speed of classification problems?

I commonly see two kinds of speed comparison for learning algorithms:

1. Algorithm A¹ converges faster than A² over some broad class of problems.

2. Algorithm A converges faster when the examples V¹ have some property than V² that lacks it.

I'm interested in the more problem-specific question (less asymptotics), here mostly specific to classification:

• Algorithm A converges faster for the class labeling L¹ of the examples V than with the labeling L^2.

I know some things that can matter for this kind of classification difficulty (e.g., linear separability of the classes, distance between their centroids, etc.), but I'd like to have better access to existing results.

Is there some useful terminology I could use for looking for these kinds of comparisons? Useful sources? Other things to know?