Apparently most people think there is some big meter attached to the network that measures the global hash. This is not true at all.

When you go to websites that tell you the global hashrates, they are giving you the information based upon a calculation. The look at the difficulty (a known number) and the time between solved blocks (a known number) and they calculate an approximation of the hash rate required to perform that. They average these numbers over several timeframes to smooth out the curves and give us reasonable approximations... but these are of course just approximations and can be skewed by large difficulty increases.

The global hash does not drop when difficulty increases, it is just a temporary lag where calculations are not reflecting reality quite so closely.

One of the contributing factors to this apparent loss of hash rates we are experiencing is the huge volume of transactions recently. The number of transactions is increasing faster than the actual hash rate. Additionally, as the global hash increases, it becomes more difficult to ship or manufacture enough hash to make a percentage increase.  You cannot have exponential hash increases forever, because if you could we would need the entire energy output of the sun by 2016 to increase the global hash by another 50%. It is slowing because it has to. 40% or 50% increases every two weeks simply cannot continue. There just is not enough energy available. I saw somewhere a chart where someone projected that if we kept increasing the hash rates by 50% every 2 weeks and had 28nm chips that by 2015 we would be using the entire energy output of the united states just for hashing.