In previous studies, we have explored the ansatz that the volume elements o
f the Bures metrics over quantum systems might serve as prior distributions
, in analogy with the (classical) Bayesian role of the volume elements ("Je
ffreys' priors") of Fisher information metrics. Continuing this work, we ob
tain exact Bures prior probabilities that the members of certain low dimens
ional subsets of the fifteen-dimensional convex set of 4 x 4 density matric
es are separable or classically correlated. The main analytical tools emplo
yed are symbolic integration and a formula of Dittmann (J. Phys. A 32, 2663
(1999)) for Bures metric tensors. This study complements an earlier one (J
. Phys. A 32, 5261 (1999)) in which numerical (randomization) - but not int
egration - methods were used to estimate Bures separability probabilities f
or unrestricted 4 x 4 and 6 x 6 density matrices. The exact values adduced
here for pairs of quantum bits (qubits), typically, well exceed the estimat
e (approximate to 0.1) there, but this disparity may be attributable to our
focus on special low-dimensional subsets. Quite remarkably, for the q =1 a
nd q = 1/2 states inferred using the principle of maximum nonadditive (Tsal
lis) entropy, the Bures probabilities of separability are both equal to roo
t2 - 1. For the Werner qubit-qutrit and qutrit-qutrit states, the probabili
ties are vanishingly small, while in the qubit-qubit case it is 1/4.