There is strong evidence for the belief that `almost all' finite semigroups, whether we consider multiplication operations on a fixed set or their isomorphism classes, are nilpotent of index 3 (3-nilpotent for short). The only known method for counting all semigroups of given order is exhaustive testing, but formulae exist for the numbers of 3-nilpotent ones, and it is also known that `almost all' of these are rigid (have only trivial automorphism). Here we express the number of distinct 3-nilpotent semigroup operations on a fixed set of cardinality $n$ as a sum of Stirling numbers, and provide a new expression for the number of isomorphism classes of 3-nilpotent semigroups of cardinality $n$. We introduce a notion of semirigidity for semigroups (as a generalization of rigidity) and find computationally tractable formulae giving an upper bound for the number of pairwise non-isomorphic semirigid 3-nilpotent semigroups, and thus an improved lower bound for the number of all 3-nilpotent semigroups up to isomorphism. Analogous formulae are also developed for isomorphism classes such as commutative and self-dual semigroups, and for equivalence classes (isomorphic or anti-isomorphic). The method relies on an application of the theory of orbit counting in permutation group actions. Our main results are accompanied by tables containing values of these numbers and bounds up to $n=10$ with computations carried out in GAP (but perfectly feasible well beyond this value of $n$).