Hanging with NVRAM in?

I've got a bunch of SLC and ELC boards, saved from back in the day,
with a handful of the connector boards they plug into.

I got them out recently because I wanted to test-boot something on a
sun4c machine.  I found I had one MK48T02B, which of course had a
long-dead battery.  So I moved it from machine to machine.  All worked,
except of course that the dead battery meant it didn't retain anything
over a power-off.

Yesterday, I found two more ELC boards, so I tried to set them up.

And I saw some very peculiar behaviour.

With the NVRAM socket empty, the boards power up and go through
selftest just fine - well, as fine as can be expected; they complain
about the NVRAM not retaining data, but that's no surprise when there
isn't any NVRAM there.  They even boot diskfully, albeit with the
00:00:00:00:00:00 MAC I mentioned recently on tech-net.

But with the NVRAM in the socket?  They don't print any of the selftest
messages on power-up, not even the first few that they print even with
no RAM installed.  (I didn't wait long - several seconds - but the
first messages are normally close enough to instant that at the human
layer I can't see any delay.)

So I tried the other boards, the ones that worked (with the same
MK48T02B) before.  Same syndrome.

But pull the chip, and they go back to "working".

Anyone have any idea what might be behind this?

/~\ The ASCII				  Mouse
\ / Ribbon Campaign
 X  Against HTML		[email protected]
/ \ Email!	     7D C8 61 52 5D E7 2D 39  4E F1 31 3E E8 B3 27 4B