We're testing IDS 11.70.FC3 on RHEL5.7/2.6.18 for a data warehouse
application, and had to turn read-ahead off due to the instance-
crashing bug related to RA. But now the project is reporting that
their ETL routine has doubled in duration. However, release FC4 fixes
the RA bug. Does Informix use read-ahead for inserts? If so, an
upgrade may fix the reported performance problem, too.

RA (Read Ahead) is primarily for reading ahead if you are doing
scans. You probably have some other issue(s) slowing down your
inserts. Follow the bytes from source to destination: Are the
reading disks slow? The network slow? The writing disks slow? Etc.
Anything different from the way you did it on the old system? Try
running "vmstat" or "sar" during the slow inserts to see if anything
is bad (IO waits, swapping, etc). Lastly, look at ONCONFIG.

-LS