Thanks but that's not actually true. If we were dealing with a non-random system with 19 possible outcomes it would br too small a sample. As everyone has pointed out this is a random system. Thus we can analysis this sample size within the following two assumptions:
1). The gear drops randomly along class line, making each class a nice simple 0.1 probability. This is what most people assume happens.
2). The gear drops randomly from a gear pool, not character based. Thus each character has 1-3 sets or a total pool of 19 sets. This makes the chance each set item drops about 0.0526.
If we look at the data provided assumption 1) fails totally. Three classes as a spike is about 0.02 likely given the sample size. The second assumption is more within our sample, only two classes fall more than 0.33 outside their predicted values. Paladins and Deathknights. Priests fall an a reason chance of occuring given their smaller size but DK's and pally's are way out.
If it turns out the second assumption is how blizzard is distributing gear then that is a weighted or bias set along character lines. So again anyone got any other info, I assume guilds that use DKP or "suicide kings" style of loot sharing can provide information there.