I think the term is "reversion to the mean".
Honestly not sure why the author chose these timeframes because they do nothing but add confusion. A 30 day, 7 day, and 1 day graph would be more useful. Maybe add in a "since last dividend" and "since start of mining" column.
ASICMiner's historical hashrate (all blocks over all time since the start of mining) has only averaged ~30 TH/s. This lifetime average has flatlined (growth slowed to a crawl) over the last couple weeks. Not sure why but, you only get paid on what is actually mined.
An all-time average will always slow to a crawl. The longer the period is the less of an impact newly found blocks will have. I included it in my chart in the beginning, when there wasn't enough data for a 7-day average, I'm thinking of removing it now because it isn't very useful.
It shouldn't be slowing to a crawl at ~30TH/s when the farm is capable of 50 TH/s. If for the prior month the network operated at exactly 50 TH/s the lifetime rate would asymptotically approach (but never reach) 50 TH/s. When it is going horizontal at 30 TH/s after months of climbing and actually declined (so each day has enough of an effect to reduce the lifetime average) that indicates something is keeping the farm from reaching its theoretical performance. Downtime, outages, orphans, etc, something.
In other words the dividends are based on an effective 30 TH/s farm not a 50 TH/s one. The shares can't pay more dividends (excluding hardware sales) then what are earned and all luck aside the network has to date operated as a 30 TH/s one not a 50 TH/s one.
Note I have faith in friedcat and am still long but to say it has no value is silly. One would expect it to asymptotically approach 50 TH/s. The fact that is hasn't (at least to date) provides insight into the long running efficiency of the network.