Errrmmm... that's a requirement not seen very often... 😉
One option would be to create a c.u.r.s.o.r. on a rather large table (to "avoid" a set based operation).
Inside the c.u.r.s.o.r.:
Join a large number of (large) tables on predicates that are not supported by any index (to force a table scan). Take as many columns as you like and build a large comma delimited string. Insert results into a table variable and use a loop to split the string again by stepping through each character.
Most important: make sure not to run it on a production system!