postgres: will not recognise existing tables

I've run into this problem twice in the last few days.

postgresql is running, shutdown, reboot, restart, and tables are missing.

The shutdown is clean:
LOG:  unexpected EOF on client connection
LOG:  received smart shutdown request
LOG:  shutting down
LOG:  database system is shut down
LOG:  database system was shut down at 2007-10-06 18:50:33 PDT
LOG:  checkpoint record is at 2/3310F20C
LOG:  redo record is at 2/3310F20C; undo record is at 0/0; shutdown TRUE
LOG:  next transaction ID: 228741; next OID: 16449
LOG:  next MultiXactId: 1; next MultiXactOffset: 0
LOG:  database system is ready
LOG:  transaction ID wrap limit is 2147484146, limited by database
"postgres"
LOG:  received smart shutdown request
LOG:  shutting down
LOG:  database system is shut down

The startup is clean:
FATAL:  lock file "postmaster.pid" already exists
HINT:  Is another postmaster (PID 99) running in data directory
"/home/postgres/
db"?
LOG:  database system was shut down at 2007-10-06 18:50:44 PDT
LOG:  checkpoint record is at 2/3310F250
LOG:  redo record is at 2/3310F250; undo record is at 0/0; shutdown TRUE
LOG:  next transaction ID: 228741; next OID: 16449
LOG:  next MultiXactId: 1; next MultiXactOffset: 0
LOG:  database system is ready
LOG:  transaction ID wrap limit is 2147484146, limited by database
"postgres"
LOG:  received SIGHUP, reloading configuration files

but then I discover a tabel is gone:
ERROR:  relation "ifl" does not exist
LOG:  unexpected EOF on client connection

All the disk space is still used, so the data is there...
I try to do a dump, but now I get:
ERROR:  out of memory
DETAIL:  Failed on request of size 20.

Ok, this I can fix, put the correct "ulimit -d" in my postgresql
startup script and it goes away - for now.

Version: psql (PostgreSQL) 8.1.3

Questions.
1) At some point I expect the ulimit solution is going to stop
working.  What do I need to do to stop postgresql from
returning this error?  Is it a shread memory setting?
Can I change something in a .conf file?
Should I be thinking about changing the way the schemas
are designed to put relations in different schemas rather
than all being in the same one?

The database only has about 8 million entries (going by a
"wc -l" of the dump'd file.)

2) Why has it forgotten about my tables?
How can I encourage it to not forget about them?

The only solution I currently have to (2) is to wipe out the
database (rm -rf) and restore from an earlier dump.  I am
slowly getting tired of doing this.

Will upgrading help?

Thanks,
Darren