Epetra solver with ilu preconditioner fails for repeated solves

The ilu preconditioner of Epetra works just fine if the solver is used just once, but fails for repeated solves, like in a timeloop.

Reconstruction:

from dolfin import *
parameters['linear_algebra_backend'] = "Epetra"
mesh = UnitSquare(16, 16)
V = FunctionSpace(mesh, 'CG', 1)
u = TrialFunction(V)
v = TestFunction(V)
A = assemble(inner(grad(u), grad(v))*dx)
bc = DirichletBC(V, Constant(0), DomainBoundary())
source = interpolate(Expression('exp(-pow(x[0]-0.5, 2)-pow(x[1]-0.5, 2))'), V)
b = assemble(v*source*dx)
bc.apply(A, b)
solver = KrylovSolver('gmres', 'ilu')
solver.parameters['preconditioner']['reuse'] = True
solver.parameters['monitor_convergence'] = True
u = Function(V)
for i in range(2):
    solver.solve(A, u.vector(), b)

The first solve is just fine, but the second fails. It does not matter whether the preconditioner is reused or not. The output using dolfin-dev on bigblue is:

  *******************************************************
  ***** Problem: Epetra::CrsMatrix
  ***** Preconditioned GMRES solution
  ***** Ifpack_AdditiveSchwarz, ov = 0, local solver =
  ***** `IFPACK ILU (fill=0, relax=0.000000, athr=0.000000, rthr=1.000000)'
  ***** rcm reord, Condition number estimate = 1.70703
  ***** No scaling
  *******************************************************

                iter: 0 residual = 1.000000e+00
                iter: 1 residual = 7.108325e-01
                iter: 2 residual = 3.981243e-01
                iter: 3 residual = 1.574911e-01
                iter: 4 residual = 3.792059e-02
                iter: 5 residual = 9.715641e-03
                iter: 6 residual = 3.437766e-03
                iter: 7 residual = 9.689458e-04
                iter: 8 residual = 2.467065e-04
                iter: 9 residual = 5.568837e-05
                iter: 10 residual = 2.356537e-05
                iter: 11 residual = 1.144578e-05
                iter: 12 residual = 2.627463e-06
                iter: 13 residual = 3.757890e-07

  Solution time: 0.081843 (sec.)
  total iterations: 13
Epetra (AztecOO) Krylov solver (gmres, ILU) converged in 13 iterations.

  *******************************************************
  ***** Problem: Epetra::CrsMatrix
  ***** Preconditioned GMRES solution
  *****
  ***** No scaling
  *******************************************************
AZ_check_options: ERROR: options[AZ_precond] has improper value = 31626240

  Solution time: 0.000012 (sec.)
  total iterations: 0
Traceback (most recent call last):
  File "simple_test.py", line 17, in <module>
    solver.solve(A, u.vector(), b)
RuntimeError:

*** -------------------------------------------------------------------------
*** DOLFIN encountered an error. If you are not able to resolve this issue
*** using the information listed below, you can ask for help at
***
*** https://answers.launchpad.net/dolfin
***
*** Remember to include the error message listed below and, if possible,
*** include a *minimal* running example to reproduce the error.
***
*** -------------------------------------------------------------------------
*** Error: Unable to solve linear system using Epetra Krylov solver.
*** Reason: Solution failed to converge (error code 1).
*** Where: This error was encountered inside EpetraKrylovSolver.cpp.
*** -------------------------------------------------------------------------