John, I never did create a patch for this. The offending line for postgres is in roundup / backends / 
back_postgresql.py, of course, and is on line 297:

sql = "select setval('_%s_ids', %s) from dual"%(classname, int(setid))

Changing this line to 

sql = "select setval('_%s_ids', %s, false) from dual"%(classname, int(setid))

fixes this problem.

I don't claim to understand the working of the ids table used to simulate sequences in MySQL. I don't 
know if MySQL has the same problem or not.

Right. After the import:

  * The IDs of the existing issues remains the same.

  * There will be an ID gap after the next issue create. 

This probably happens with other types of objects.