Can SERVFAIL be incorrectly returned through caching?

[Simon Waters]

A BIND 9.2.1 server on Redhat Linux 7.0 built from source,
default options.

After a brief period of being disconnected, and following a
reconnect a dig to the local nameserver for a record returned
SERVFAIL.

This continued after the name server was reconnected for a
noticable period.

The SERVFAIL error was reported long enough for me to manually
follow the resolution procedure down from the root name servers,
and confirm that nothing obvious was wrong (routing/DNS
delegation).

So course of events....

disconnect from internet.
Look up record - SERVFAIL
Reconnect to internet
Look up record - SERVFAIL
Lookup up other stuff appear okay.
Lookup uk NS - okay
Query co.uk NS for ISP NS - okay
Lookup ISP NS - okay
Query ISP NS for record - okay
Lookup record - SERVFAIL
Lookup record - okay

The record was "news.demon.co.uk" but I'm more interested in why
I might get a SERVFAIL once connectivity is restored. Is there
an additional caching going on I'm not aware of? I'm not
particularly concerned by the episode other than I can't explain
what the order of what happened.

The whole issue resolved itself quickly and I guess I can't rule
out a transient network problem dropping the odd packet, or a
change at the ISP (although this seems unlikely).

No LAME server messages were logged, and whilst some odd packets
were dropped during the episode these appear to be related to
some overkeen Microsoft PC somewhere wanting to talk NetBIOS to
my nameserver.