Performance difference for CTMC model checking

[volkm]

I get different runtimes for CTMC modelchecking when computing the transient probability of reaching a label compared to avoiding the label.

I attached a minimal CTMC modeling a phase-type distribution with 9 states. I compute the following properties:

1. P=? [F=30000 "goal"]
2. P=? [F=30000 !"goal"]

Model checking the first property is over 10 times slower than the second property even though the second one is just the inverse of the first one.
The complete output is:

Command line arguments: -drn ctmc.drn --prop 'P=? [F=30000 "goal"];P=? [F=30000 !"goal"]'
Current working directory: ~/storm/build

Time for model construction: 0.000s.

-------------------------------------------------------------- 
Model type: 	CTMC (sparse)
States: 	9
Transitions: 	9
Reward Models:  none
State Labels: 	2 labels
   * init -> 1 item(s)
   * goal -> 1 item(s)
Choice Labels: 	none
-------------------------------------------------------------- 

Model checking property "1": P=? [true U[30000, 30000] "goal"] ...
Result (for initial states): 0.6646109497
Time for model checking: 0.018s.

Model checking property "2": P=? [true U[30000, 30000] !("goal")] ...
Result (for initial states): 0.3353890503
Time for model checking: 0.000s.

Do we have an idea what might be responsible for this difference?