There was no udump trace created.
I went ahead and explicitly started a SQL trace and reproduced the issue. Interestingly, while the slowdown is still a major issue as progress essentially stopped at about 2600 records fetched, the connection did not drop.
I profiled the waits on the session...
We created a 10.1.0.5 database (64-bit database running on SUSE LINUX Enterprise Server 9 x86_64), and now we have all plsqldev clients in the network experiencing the same issue. Versions include 7.0.3.1123 and 6.0.5.931.
For a sufficiently wide table (more than 3 or 4 columns) of only native...