Ah, OK. If data grouped in table2 is significantly complex I'd expect a little performance gain. But with an efficient DBMS I'd believe that the overhead added by joining the tables should lose the gain from memory saving.

BTW, 4 days to build a unique index seems to me really too much! I have a table with about 10 million rows (but only 6 columns, I've to admit) that build its primary key (on two columns) in about an hour. And last time I had to restore it from a backup I grumbled for it was too slow... :)