I don't think the sample of experiments reviewed is large enough to evaluate sample size versus effect size; throw out the outliers and there's nothing left.

The first Rosenthal meta-analysis used 345 studies. That is pretty big. And the individual studies listed in table 17.1 have large n, ranging from 79 to 5000+.

I'm now heavily concerned about the validity of the IQ test used; however, that's more due to the 8 point increase in the control group, when no increase is expected.

No, that's not a problem that should concern you. Children IQ scores are less stable than older people's scores, test-retest effects will give you a number of IQ points (that's why one uses controls), and children are constantly growing.

What should concern you is that the researchers involved were willing to pass on and champion a result driven solely by obviously impossible nonsensical meaningless data. A kid going from 18 IQ to 122? or 113 to 211? This can't even be explained by incompetence in failing to exclude scores from kids refusing to cooperate, because tests in general (much less the specific test they used!) are never normed from 18 to 211. (How do you get a sample big enough to norm as high as 7.4 standard deviations?)

Worrying about the control's gains and not the actual data is like reading a physics paper reporting that they measured the speed of several neutrinos at 50 hogsheads per milifortnight, and saying 'Hm, yes, but are they sure they properly corrected for GPS clock skew and did accurately record the flight time of their control photons?"