A good starting point might be this 2004 review by Schmidt & Hunter. It tabulates correlations of 0.23 to 0.58 between "general mental ability" and "supervisory ratings of job performance", with the correlation increasing monotonically with a five-level rating of job complexity.

Edit: I should probably add Schmidt & Hunter's note that those averages "are corrected for measurement error in the dependent variable and for range restriction but not for measurement error in the GMA measure; hence, these are estimates of operational validities, not construct-level correlations."