POR strangeness on Cyclone III

[Altera_Forum]

So here's a wacky one. I'm having problems with a Cyclone III design I'm working on; the power-on reset logic (as inferred by what I'm seeing on nSTATUS) has radically different behavior the very first power-on than it does subsequently. 

 I'm doing a slow PS configuration; I bring nCONFIG high with the power-on, then after 300 ms I blip it low for 1 ms then back high, then wait 1 ms, then start pumping configuration data. Normally nSTATUS comes up 100 ms after the power rails are stable, as expected; 200 ms before my attempted configuration and all is fine.  But on the very first power-on of the unit, nSTATUS doesn't come up at 100ms. Instead it stays low for about 700 ms, then chatters up and down, then eventually goes high at around 850 ms. Needless to say, this messes with my configuration plans. See the attached picture; yellow trace is nCONFIG (300 ms past power-on) and blue trace is nSTATUS.  The power rails come up monotonically and in under 50 ms. nSTATUS is pulled up to 3.3V with 10K. We tried speeding up the power rail ramp-up to only a couple of ms, this eliminated the chatter but nSTATUS still doesn't come up until the 850 ms mark.  And then the second power-on, it's fine. And the third, and forth and fifth. And you can unplug it for a week, or a month, and it's fine. But if you leave it unplugged for several months, the problem will recur. We've been able to speed this up by leaving the unit in a 120C oven overnight, that seems to consistently give us our one recurrence of the problem before we have to cook it again. But it's not just some weird assembly problem; we're seeing it on 100% of the units off the line. And it's not the power supplies doing something different; we've checked and the power-on transients look identical.  I've checked the C3 errata, and don't see anything. C3 is a pretty mature technology at this point; strange power-on glitches should have been found by now. Anyone got any thoughts at all?