After few hours of host outage MW project started to refuse sending work with messages about BOINC running only 80% of time.
With my cache settings, MW's project share, allowed MW task queue and completion time for MW task on my host it's absolutely nonsense to think MW tasks can't be finished in time.
http://milkyway.cs.rpi.edu/milkyway/forum_thread.php?id=952&nowrap=true#28115

ADDON: even by BOINc's estimation time to complete was never bigger than 1,5 hours.
MW deadline is 3 or 4 days. Host's cache didn't overloaded. Why MW server refuses to give work in such situation?

ADDON2:
It seems the same issue as here:
http://boinc.berkeley.edu/dev/forum_thread.php?id=4079

---

For the same reason as is discussed in this thread, it is a bug in the 6.6.3x versions where the delay_time is calculated wrong.

A fix is in the new source code. You can build a new client using that source code or wait for Berkeley to release a new test client. I have built a client with the new code and it works quite well. It can be found linked to in that same thread.

---

Thanks!
Unfortunately, I running modded BOINC with per-project network suspending feature (in attempt to help SETI with upload problems).
Does this fix already incorporate fix about upload policy change (so I can get both and build BOINC that will work as intended at last) ?

---

Does this fix already incorporate fix about upload policy change (so I can get both and build BOINC that will work as intended at last) ?

Yes, and working fine I saw. I had 1 result in upload mode earlier and 4 sitting there "pending upload".

---