[Samba] Excel file corruption - IRIX 6.5.14m and Samba 2.2.2

[Scott Merrill]

I have recently encountered a problem with Samba.  While researching it, I 
found this post:
http://lists.samba.org/pipermail/samba/2002-March/068091.html
(Thanks for the reply, Jijo!)

Although similar, my situation is a little different.  The following steps 
produce the same result every time across multiple workstations:
Create a new Excel file.
Enter some data.
Click "FILE", "SAVE AS".
Save the file as H:\test1.xls.
 (where H: is the user's home directory on the server)
Close the file.
Open H:\test1.xls.
Modify it (or not, it doesn't matter).
Click "FILE", "SAVE AS".
Save the file as H:\test1.xls.

At this point, a dialog box pops up stating that the file has been saved, but 
the workstation is low on memory.

Close the file.
Open H:\test1.xls.

Another dialog box pops up, stating that the file is not in a recognizable 
format.

I realize it's an odd sequence -- how many times do you want to SAVE AS a file 
to itself? -- but the fact that users can easily corrupt their data is an 
issue.  Simply choosing SAVE does _not_ corrupt the file.  It's only when you 
elect to SAVE AS using the same file name.

The server is an SGI Octane running IRIX 6.5.14m and Samba 2.2.2 (downloaded 
from freeware.sgi.com) on an XFS filesystem.  I've just downloaded the 2.2.3a 
tardist file from the Samba site, but cannot presently install it (must wait 
until after working hours).

I've looked through the logs for the workstations I executed the above steps 
on, but I don't see any oplock messages.  Do I need to increase my logging 
verbosity?
Will upgrading to 2.2.3a resolve this issue?
Will turning oplocks off, or disabling nt acl support (as per the above 
thread's suggestions) resolve this?