How many variables in your model and what is the transfer function of that model?
What resolutions are needed? Current resolutions are woefully inadequate, typically 100 km horizontal, 25 vertical layers, and 10 minute intervals. The last one, time interval may be ok, but the grid resolution clearly sucks since climate affecting weather is mesoscale 10km and sometimes small scale (1km) in the case of tropical shower activity. The essential problem is that small scale weather activities like the tropical showers pump water vapor up into the upper atmosphere where it causes the most amount of greenhouse warming. If they are not modeled accurately, then the climate prediction results will be wrong.
My point in bringing up the models was that once computer power is sufficient for adequate models, they will be useful to calculate the effects of simple and cheap climate countermeasures that might be needed. The current Kyoto countermeasure is estimated to cost about $100 billion for each 1/1000th of a degree of hypothetical cooling, an outrageous amount of money for a completely inconsequential result.