I've come across this one quite a bit at one contract I had as a third-party vendor insisted that the performance of a binary collation was much quicker than a case-insensitive one and so set the default to Latin1_General_Bin for everything.

I'm not sure if there are tests/stats to back this claim up but I do know that the majority of queries against the data were name related and so most of their queries were written as UPPER([NameColumn]) = UPPER(@Criteria). Brilliant !! Index seek to index scan in one easy step.

After seeing this I wasn't inclined to believe their performance enhancement claim.