Miners returning 71% of expected yield

[Cryptoman]

I've been mining for about 3 weeks now, so I thought I'd put together a spreadsheet (see link below) comparing expected and actual returns. It is not a trivial calculation since the difficulty keeps changing and so does my mining cluster. I created one row in the spreadsheet each time something changed (difficulty, hash rate, payout) and calculated what should have been expected returns over the previous time period. As it turns out, I initially did as much as 33% better than expected. Recently, my returns haven't been as good, and overall for the 3 weeks I'm 29% below expectation. It's a relatively short time period, so I don't necessarily think anything is wrong.  Has anyone else compared theoretical versus actual returns? Perhaps the spreadsheet will be a useful template for others.

Links
The spreadsheet: http://www.mediafire.com/?d46q1zku7u5y8fi
Generation time calculator for arbitrary difficulty factor: http://developer.wolframalpha.com/widgets/gallery/view.jsp?id=76444b3132fda0e2aca778051d776f1c
Difficulty changes: http://nullvoid.org/bitcoin/difficultiez.php

[edit] I'm running m0mchil's miner, and the hash rates in my spreadsheet are simply the sum of the rates displayed by the miner instances.

[ribuck]

The randomness of block generation follows the Poisson distribution. I'll leave you to work out the math, but 29% below expectation doesn't sound alarming.

[dust]

It is possible that your GPUs are throttling themselves due to high temperatures.  Are you using actual sustained hashrates for your calculations?

[Cryptoman]

It is possible that your GPUs are throttling themselves due to high temperatures.  Are you using actual sustained hashrates for your calculations?

One of the 5870s was throttling around 95 C because it was next to the wall of the case and wasn't getting enough cool air.  I've since remedied that.  I have another 5870 that is running a little warm (85 C), possibly because it needs more thermal grease, but it's not throttling.  I just watch the hashrates displayed by poclbm.py and note what value they level out at.  I assume these rates are sustained since the miners aren't used for anything other than mining.

At any rate, I'm not saying there's a problem since the miners have only been online 3 weeks.  I just haven't seen anyone else post a detailed comparison of actual vs. theoretical returns.

[dust]

I have had issues with 5870s jumping between 320 and 220 mhash/s at temperatures around 90c.  It only happens after the miner has been running for a while.

[Raulo]

At any rate, I'm not saying there's a problem since the miners have only been online 3 weeks.  I just haven't seen anyone else post a detailed comparison of actual vs. theoretical returns.

Unless you have problems with miners, it's just bad luck.
You did 14 blocks with average rate of success of 19.68 (if you correctly calculated it). Poisson distribution gives 11.8% probability of getting 14 blocks or less with average success rate of 19.68.
http://stattrek.com/Tables/Poisson.aspx

One every 9 persons was as unlucky as you. Those who are lucky and get more than average do not post on the Forum.
If you want less variability, use a pool.