The probability that your parameter (i.e., S) takes on a value of exactly 0.6 (i.e., 0.60000…) is zero. It makes sense to talk about the value of a random variable falling within some interval (e.g., 0.6 \pm .01 or 0.6 \pm .00001), but the probability of it taking on any one of its infinite possible values is zero.
But this leads us back toward the problem that loss functions are designed to address. If the true value of your parameter is 0.6, how bad would it be to erroneously conclude that the value was instead 0.60001? What if you erroneously conclude it’s 0.7? Or 0.01? Exclusively focusing on the inferences that yield the correct parameter value (e.g. 0.6) means that you end up ignoring all the potentially incorrect parameter values that you could instead infer, both the near misses and the ones that are way off.
If the true value of some parameter is 0.6, which of these two posteriors would you prefer your model to produce?
The former has non-zero density near 0.6. As a result, your scheme would score it “better” than the latter. With the latter, all credible values are incorrect, but pretty close.