The fix for this problem, under Redhat Linux follows: I don't know whether 
these remedies apply to other OS.

Contrary to the "Server Upgrade Guide"  run the extractdb step as root, not as 
the instance owner.

Before starting the extractdb step, set the environment variable DSMSERV_DIR to 
/opt/tivoli/tsm/upgrade/bin 

If extractdb cannot find dsmserv.dsk  copy it to .../upgrade/bin, and cd to 
.../upgrade/bin before starting the extractdb

Now it's running.

The insertdb step is reporting 62 GB per hour at this time.  It started 
somewhat faster, and has steadily decreased in its reported throughput.

ANR1379I INSERTDB: Read 45,465,996,483 bytes and inserted 373,009,536 database
entries in 0:40:00 (65038.50 megabytes per hour).
ANR1379I INSERTDB: Read 50,755,623,849 bytes and inserted 413,279,104 database
entries in 0:45:00 (64538.67 megabytes per hour).
ANR1379I INSERTDB: Read 55,960,204,550 bytes and inserted 452,175,951 database
entries in 0:50:00 (64040.40 megabytes per hour).
ANR1379I INSERTDB: Read 61,469,683,087 bytes and inserted 494,880,598 database
entries in 0:55:00 (63951.27 megabytes per hour).
ANR1379I INSERTDB: Read 66,385,342,814 bytes and inserted 534,451,025 database
entries in 1:00:00 (63309.00 megabytes per hour).
ANR1379I INSERTDB: Read 76,866,344,723 bytes and inserted 619,922,387 database
entries in 1:10:00 (62832.86 megabytes per hour).