When c is close to b-a (as in test case 4, with a=3, b=4, c=1.01) the solutions are susceptible to roundoff error; after some checking, I believe the correct answer for test case 4 (to 15 significant figures) should be A=0.00650196332827386 rather than A=0.006501963328256.