Which equation will give me meaningful insight? I have 10 iPads. I am logging the number of times an app is crashing each day for each of these iPads. The number of crashes tends to be skewed towards just a couple of the devices such that taking an average will not give anything meaningful. What type of equation should I use to get some meaningful data related to my app crashing? And / or (more importantly): What should I search for in order to learn about the type of equation(s) that will help me?
Example dataset: 
{iPad,numCrashes}
[{0,11},{1,7},{2,10},{3,0},{4,0},{5,0},{6,0},{7,0},{8,0},{9,0}]

Crashes per device is 2.8, but only 30% of the devices actually crash.
 A: As you stated in your post, 30% of the devices actually crash. Of the iPads that crash, they crash, on average, 9.3 times each day ((11+7+10)/3 = 9.3). Aside from that, there aren't any fancy statistical tests that can be used as you only have data on 10 iPads. This is simply an exploratory study that may warrant further investigation. I hope that helps!
A: I would think about it as a frequency/severity problem, which has sort of been suggested in the OP and the comments. Keep track primarily of the % of iPads that crash, and then separately of the # of crashes per iPad conditional on it crashing. Then you can start looking for other data points that might explain these two things. If one of them changes but the other doesn't, you'll have a little more understanding about the impact of any changes you may have made to the app or OS or whatever. 
A: I'll assume that each iPad had a priori the same chance of having mulfunctions (i.e. you tested them on comparable amount of apps and time). If my assumption does not hold, then you simply cannot make inference: maybe all iPads are buggy, but only the first three were actually used and properly tested (and the rest maybe never left their original manufacturer packaging)?
Anyway, the data has Poisson distribution, and it seems that you want to predict if the number of failures has anything to do with some predictor. The Poisson regression exactly answers this question. Please mind, that you will need to supply an independent variable that tries to predict the failures.
If you want to calculate some sort of significance, please be aware, that the data set is too small for safe out-of-the-box usage of any classical statistical tools. You should at least do bootstraping/permutation testing of statistical significance of fitted regression parameters. 
Or even better, do Bayesian extension analysis (i.e. Bayesian Poisson regression) which has completely different view on how to draw conclusion from empirical data, and which is proven to work also on extremely small data sets.
