Please name some actual examples in which the truth or falsity of any particular "conjecture[] in prime number theory" has a observable bearing on the performance of any algorithm.

I felt it was clear from context that the OP meant one of the serious conjectures like GRH, which is an awful long way away from being 'shaky'.