It has been shown that training effects and the FLR effect are not g-loaded. Spitz in his discussion of the many types of compensatory education that has been tried, mentions that they were often suspected of teaching to the test specifically. This shud show up as non g-loaded gains. Unfortunately, the data are probably lost now. If only we had mandatory data sharing!
There is however one option, Head Start. Since this program is STILL running, one might be able to get some data from it.
As an update to this, we did a study of it. https://emilkirkegaard.dk/en/?p=4361