[pymvpa] Time for single permutation varies between a few seconds and several minutes

[Jan Derrfuss]

Excellent thought, Michael! This does indeed appear to be the case! I 
tried to test this hypothesis by printing the permuted labels and 
creating a new attributes file with the labels from a "slow analysis", 
which I then passed to the original searchlight analysis. This analysis 
then took about the same time as the permuted one.

What I still don't quite understand though is why the times for the slow 
analyses seemed to cluster around ~ 240 s. Intuitively, I would have 
expected them to be more variable.

Thank you!
Jan