(sorry for going so far off-topic in the Eligius thread)

1) Average over how many blocks?  The graphs at Bitcoinwisdom show averages over the last 2016, 1008, and 504 blocks, as well as the last 16 blocks on a different graph.... of course the average over 2016 blocks varies less (and is ultimately what determines the difficulty increase), but the average over 1008 or 504 blocks might be more reflective of trends in hashrate.  If you are averaging ONLY over the blocks since the last difficulty increase, you're throwing out useful information, because....

2) Taking the average time to solve (however you determine that average), and doing some math with the current difficulty level, lets you back into an estimate of network hashrate.  This is the more stable number since of course the time to solve will change at every difficulty reset.  (In other words, the hashpower is more stable versus the block time.... in the real world, the block time is determined by the network hashpower, but since we have no way of knowing the true hashpower of the network due to its decentralization, we have to work backwards mathematically, and derive the hashpower from the block time.)  The bitcoinwisdom graphs show that the network has reached roughly 300,000 TH/s several times on both the 504-block and 1008-block lines, and come close on the 2016-block line.

So the upshot is that you can and should use the block times prior to the latest difficulty reset. By doing calculations based on the difficulty at that earlier time, you can estimate the network hashrate, which has a tendency to monotonically increase (i.e., it may go up or stay flat, but it will rarely go down for any sustained period).