I'll assume that you've already asked yourself the question about "should you combine the utilities into one table" and answered in the affirmative. Of course there are lots of worries about doing this and some potential for overcounting the value of some attributes and undercounting the importance of others, but, again, we'll assume that you've considered those worries and judged that they don't apply to your survey.

In this case, you could go back and rerun the CBC analysis, including all the choice sets in a single analysis. You would go back and recode the data matrix so that you had 0 codes (for "missing") for attributes E, F and G when you were modeling choices from the first study and then 0 codes for A, B and C when you were modeling choices from your second study, If you concatenate the two data matrices and run them as a single model, well, there's your utilities, all in one model.

If you don't have access to the raw data, a very simple thing (not ideal, because better would be combining the data into a single model as described above) to do would be to use the cost utilities from study 1 to predict the cost utilities in study 2. For example, if you have 5 levels of cost, then in study 1 your 5 cost utilities are x1, x2, x3, x4 and x5 and y1-y5 are your cost levels for study two. Now run a simple linear regression analysis to predict your y variable with your x variable and voila, you have an equation that calibrates your study 1 utiliteis to fit with your study 2 utilities. This is one low tech way of "bridging" utilities between two sub-experiments.