With the constant difficulty during that time and the 2250 GH figure I used for hashing power I figured they should have gotten 133 solved blocks, and solved a block every 53m47s. They only got 105 blocks. That's 28 missing blocks, or $19,000. About 807 blocks went by in that time. That number is important because you need to know the number of trials. Probability (p) of finding a block is 0.165 (that's 133/807).
If we assume that you are correct on expectation of 133 while only 105 blocks found (I have not checked it) than probability of finding at most 105 blocks out of expected 133 is 0.6952%. This is rather high certainty i.e. more than 99.3% that something is badly wrong there. I'd suggest anyone to jump the ship ASAP (I would).
Check out
http://mining.mainframe.nl my math shows that no cheating going on there.
On longer time period it is very easy to spot cheating. Take as long period as you have data for. Calculate expected number of blocks solved, calculate actual number of blocks solved. It should converge. The law of large numbers is working. A few such bad streaks as described above and certainty that a pool is cheating quickly goes over 99.99%.