Interesting. I'd be curious to see the query plan on the bigint implementation, and a link to the UDT implementation if anyone has one (though I agree the overhead in storage is substantial).