Scale separation and statistical pseudoequilibrium (forget the exact term) are just assumed in order to "get results". I.e., if you didn't make those assumptions, simulations would require so many data points that they would be impossible. Nonphysical energy damping is added, since without it, most simulations blow up. Lots of important parameters are just picked roughly at random, just to calibrate the model to historical data.
Stock market predictions also tend to have one major advantage over GCM models - their proponents admit their limits.
If you'd like to see a chart summarizing the various climate forcings, then look at page 4 of this link: http://www.ipcc.ch/pdf/assessment-report/ar4/wg1/ar4-wg1-spm...
I'm not going to track down all the literature behind those numbers here, but there are published, falsifiable arguments supporting them. Climate models aren't similar to Wall Street voodoo.
And this is simply the measurement part. _Predicting_ what will happen is a thousand times more difficult. If you want predictive science based on the laws of physics, look at astronomy. 4000 years ago, the Babylonians could predict astronomical events with much more precision than anything today's climatologists can predict about average temperatures.