Employing Hilbert-Schmidt measure, we explicitly compute and analyze a number of determinantal product (bivariate) moments |rho|^k |rho^{PT}|^n, k,n=0,1,2,3,..., PT denoting partial transpose, for both generic (9-dimensional) two-rebit (alpha = 1/2) and generic (15-dimensional) two-qubit (alpha=1) density matrices rho. The results are, then, incorporated by Dunkl into a general formula (Appendix D6), parameterized by k, n and alpha, with the case alpha=2, presumptively corresponding to generic (27-dimensional) quaternionic systems. Holding the Dyson-index-like parameter alpha fixed, the induced univariate moments (|rho| |rho^{PT}|)^n and |rho^{PT}|^n are inputted into a Legendre-polynomial-based (least-squares) probability-distribution reconstruction algorithm of Provost (Mathematica J., 9, 727 (2005)), yielding alpha-specific separability probability estimates. Since, as the number of inputted moments grows, estimates based on |rho| |rho^{PT}| strongly decrease, while ones employing |rho^{PT}| strongly increase (and converge faster), the gaps between upper and lower estimates diminish, yielding sharper and sharper bounds. Remarkably, for alpha = 2, with the use of 2,325 moments, a separability-probability lower-bound 0.999999987 as large as 26/323 = 0.0804954 is found. For alpha=1, based on 2,415 moments, a lower bound results that is 0.999997066 times as large as 8/33 = 0.242424, a (simpler still) fractional value that had previously been conjectured (J. Phys. A, 40, 14279 (2007)). Furthermore, for alpha = 1/2, employing 3,310 moments, the lower bound is 0.999955 times as large as 29/64 = 0.453125, a rational value previously considered (J. Phys. A, 43, 195302 (2010)).