That is what they started with but they had locking and blocking issues while updating that one nextnum table to obtain the next key value.
I think I was unclear.
I am suggesting that instead of one dummy table for each key-value, you create a single dummy table and grab the next identity value from it just as in your solution. The only difference is that the dummy table is shared amongst all of the key value pairs. This may, of course, leave gaps in the sequence for each key-value but you have stated that that doesn't matter in this case.
--
JimFive