Based on Nikolai's logs, I have identified this as a very old bug that seems to have only caused block production failure in rare cases. It will be fixed in 0.6.0.
I can explain the details of the bug after 0.6.0 is released, but until then there is likely not much risk. If large numbers of delegates do happen to freeze before that happens, they can temporarily work around it by doing:
And the default setting can be restored as: