matrix%matrixType, & !Matrix type
matrix%numRows, & !Dimension
matrix%firstInRow, & !Row cummulative
matrix%columns, & !Column indices
If you suspect that the solution x in your "first case" contains incorrect values, you should check whether the matrix A and the vector b have been given correct values.
For the data that you supplied for "case 1", I find that there are only two items in A.x - b that are larger than 1E-11 in absolute value, and none that exced 1E-10.
Thus, MKL/Pardiso is giving you a reasonable solution, but you have to look into the question of whether the matrix is well-conditioned or not.