Performance question about "next number" methods

Can anyone comment on the performance of the following alternative "next number" techniques:

1) increment a value stored in a data area (e.g. in RPG, something like this):

*LOCK IN DATAAREA
ADD 1 NEXTNBR
OUT DATAAREA

2) store a record in a keyed "next numbers" file that stores the current value for each file
(where the "key" is the name of the file that this next number pertains to)

pseudo-code below:
read record with key, with a record lock
increment next number value
update record, releasing the record lock
return new value to caller

3) using newer features in DB2 UDB, e.g. a sequence or identity field

These techniques could be implemented in any ILE language, RPG IV, COBOL, C, etc.

I am interested in the relative performance of one technique versus another.

Thanks.

Mark S. Waterbury