Fluid and openMP


I made some experiments with fluid simulator noticing some strange behaviour with openMP. Starting blender and running a very simple simulation (fluid cube inside a cubic domain) all processors are at 100%.
Then stopping and running again (with no parameters change) all processors are at 40-50%.
Closing and restarting blender, opening the previous .blend, rerunning the simulation, again all processors run at 100%.:spin:

I’m using the last revision (33504) and I notice this behaviour with Quad, i7 (HyperThread enabled and disabled) and Linux (Fedora 13, Ubuntu 10.10).

It seems that after running a simulation something of openMP remains appended somewhere, is this correct or a bug? I read in some threads of this forum about the partial load of processor but in this case the first simulation is ok.