The work unit you have seen is a pre-run work unit that was mentioned earlier in the thread. We run about 3 of these per batch of 5000 work units. They are sent to reliable hosts only at high priority which is why the shorter deadline. This allows us to catch if the estimation script gets way off for a batch. If these pre-runs have issues then we are able to rebuild them before the bulk of the work units are released. You get normal credit for these and these results are still sent back to the researchers.

Hope this helps,
-Uplinger