Justify Coding Gain Claim

Over on David Rowe's blog post "FSK over FM", there's a claim that the relative coding gain for 2-FSK vs 2-AFSK is approximately 7dB for weak signals. In the CATS document, the claim is 12dB. Is there a citation and/or relevant analysis to support the 12dB claim?

Related pedantic question, is this a "coding gain", or a "modulation gain"? Seems like the latter, whereas the FEC/LDPC is the former?