While I agree you should not make rash decisions and jump out of your pool based on your payout dipping due to variance. There does seem to be some suspicious patterns occuring.
Mining Pools on the other hand typically calculate their hash rate based on the number of shares turned in. For example:
https://www.btcguild.com/index.php?page=pool_stats and
http://eligius.st/~gateway/stats/recent-blocksInterestingly both of those pools have had an increase of their hash rate, and a drop in pool luck over the same time frame.
What if a major pool, or private miner (Ghash, CoinTerra, etc), mined with half of their hardware on a public pool but executed a withholding attack on the pool? They would collect BTC rewards for the other blocks found by the public pool, but withhold all blocks found and transfer thier BTC rewards to their own pool. This would make the public pool's luck look bad, and their own pool luck look good.
They could even delay the block release until a time when their own pool luck was bad, or the public pool's luck was good, to prevent it from looking too obvious. This delay would cause the entire combined network hash speed to appear much lower than it actually is, at least until they release the found blocks. They would only need to release them before the difficulty increased when the blocks might become invalid.
Disclaimer: I have done no research beyond the links listed above, and this is pure speculation.
Now that I see Ghash is also paying out less than expected I have to go "hmmm"...and do more research.
http://www.bitcoincharts.com/Current network speed (calculated by rate of blocks turned in): 147.9PH
https://www.btcguild.com/index.php?page=pool_stats BTCGuild reported speed (calculated by rate of shares hashed): 12.0PH
http://blockchain.info/poolsBTCGuild reported share of total network (calculated by rate of blocks turned in): 7%
149.9 * .07 = 10.4PH = 12PH - 1.6PH missing blocks
Now of course this can be explained by bad luck, but on how many pools can this bad luck occur.
Also note that the share of network blocks turned in is 9% for two individual unidentified addresses, and 12% completely unknown to blockchain.info.
Obviously there are private players out there, but are they 21% of the actual hashing capacity?
Last link:
http://bitcoin.sipa.be/speed-lin-10k.pngA lot of blocks sure are being turned in now that we're in the last 300 blocks before the difficulty change.