There seems to be the misconception that faster hardware could generate more coins. But this is only true in so far as you might be able to get a bigger share of the fixed daily production of 50 * 6 * 24 coins that are generated on average. Lets say I would have a super-computer which could generate 20% of the current bitcoin network procossing power in Hashes per second.. but the cost for operating it would be much higher than what the expected amount of solved blocks/bitcoins would cost at an exchange. Then I would not power up my super node but start buying BTCs from a cheaper producer.

Currently the electricity costs for mining a BTC are lower than the price payed at exchanges like Mt. Gox. But already today the profit margin is different for systems with different efficiency rates and also depending on the price for electricify which a 'legal' miner has to pay. We will very likely see in the next couple of month if not faster, that the profit margin for current GPU hardware will go towards zero. And the important point here is that it is not possible to buy a faster GPU that is less energy-efficient to increase the profit margin - if you cannot improve the efficiency of your system or find a 'provider' electric power who is cheaper .. you would loose money continuing to mine with your inefficient (sub-threshold) system.

Therefore: Speed alone (hashrate) does not help you. A 'legal' miner can only make profit (and e.g. generate the money for his/her rig) when he is using hardware that has a kW/h per GHash efficiency above a certain threshold.

Viewed differently: Imagine the daily 'production' of ~7200 coins would be worth 100.000 USD at Mt. Gox and that price would stay constant. At the same time the number of miners would increase until a saturation plateau is reached. If all miners would daily convert their coins (not acting as traders/speculator) an ever increasing fraction of the 100.000 USD would go to the suppliers of electric energy...