@ Comment 14 by vanghelie:

Who is sacrificing something so that IT has less chance of producing offspring? The gene results in some workers sacrificing something for other workers and (directly or indirectly) the welfare of the queen - but they have 0 (or very little) chance of producing offspring anyway. As far as the queen is concerned (which is the only reproducing agent here), the gene only results in an increase of effectiveness of how she is "cared for" (regardless if some workers suffer for others because of it..)

OK. I think I see your question. If the ants had already evolved the sterile worker-caste system (which requires hamilton's equation because the workers are sacrifice reproductive health) then subsequent details of evolution of features of the worker caste do not represent further sacrifice, as you point out. But that just means "C" in the cost side of rB>C is zero for those new traits like soldier jaws. B is the benefit to the queen's reproductive success, and r is the relatedness of the worker to the queen. So both the cocept of kin-selection and Hamilton equation are relevant, but C=0 (or almost zero) may make Hamilton's equation seem trivial... but remember that B must both remain positive.

Bear in mind that mutations can come along any time that make workers more likely to reproduce (a "negative cost" to the worker) but the same mutation may also have "negative benefit" for the queen, and Hamilton's equation will apply in that case as well ("is rB more negative than C?").

rB>C is a kind of minimum requirement for the gene responsible to be selected. It is interesting to turn it around so that B' is benefit to the individual (B'=C) and C' is cost to kin (C'=-B), then we have B'>rC'-- a trait that benefits me personally must not cause an excessive net cost to my kin!

When it comes to competing traits (genes for large vs. small jaws, or whatever) if the genes are mutually exclusive then (r * B_1) >?< (r * B_2) is the thing that matters for the already-sterile workers. More generally, what matters among competing traits is whether (r * B_1) - C_1 >?< (r * B_2) - C_2. (There does nto seem to be a way to do subscripts... sot the _1 etc, are subscripts)