Sure enough, you can;t set the shell in a script called by the qsub script; you have to specifiy it in the qsubscript.

I now have sort of got it running:

CCSM PRESTAGE SCRIPT STARTING
- CCSM input data directory, DIN_LOC_ROOT_CSMDATA, is /work/00671/tobis/inputdata
- Case input data directory, DIN_LOC_ROOT, is /work/00671/tobis/inputdata
- Checking the existence of input datasets in DIN_LOC_ROOT
CCSM PRESTAGE SCRIPT HAS FINISHED SUCCESSFULLY
rm: No match.
Fri Dec 17 17:41:16 CST 2010 -- CSM EXECUTION BEGINS HERE
Fri Dec 17 17:41:20 CST 2010 -- CSM EXECUTION HAS FINISHED
ls: No match.
Model did not complete - no cpl.log file present - exiting
TACC: Cleaning up after job: 1731370
TACC: Done.

To be clear, I have now loaded the executable, which promptly died without leaving a clue as to why anywhere that is obvious. Of course, who knows where it thinks it ought to leave the clue. I have set "find" the task of finding files created over the weekend. It is amazingly slow, though.

This still amounts to progress: after a week I have actually got the thing to lurch to life and die.

Life in the fast lane.
...
AHA!

$WORK/CAM_3/run/ccsm.log.101217-174114

it says, 32 times,

MPI_Group_range_incl(170).........: MPI_Group_range_incl(group=0x88000000, n=1, ranges=0x48edec0, new_group=0x7fffc8173f2c) failed
MPIR_Group_check_valid_ranges(302): The 0th element of a range array ends at 31 but must be nonnegative and less than 1
MPI process terminated unexpectedly

OK, before I go whining around, I will try to redo everything.