It was in one of my small lab machines.
SQL Server 2012 SP2 Developer Edition on a VM on ESXi 5.5, 2 vCPU, (1 virtual socket, 2 virtual cores per socket), 8 GB RAM (SQL Server's max set at 5), physical host using a Xeon E5520.
I ran two tests. One populated 3 pairs of tables with 10k rows: a pair with a single TINYINT column each, a pair with a single INT column each, and a pair with a TINYINT column and a CHAR(3) column each (to match the size of the INT tables).
I then ran queries joining each pair of tables (forced a nested loop join both to prevent the comparisons from being hashes in a hash join, and to force a large number of comparisons), with the output piped to a variable to avoid SSMS overhead.
None of those queries consistently outperformed the others.
In the other test I declared a bunch of TINYINT and INT variables, and set up two loops that performed the same set of 50 math operations one million times, one loop using only the TINYINT variables, and one loop using only the INT variables.
There was no consistent difference between those either.
I was going to post the actual scripts, and might when I get some extra free time, but it seems I didn't save them so I'll have to rewrite them when I have the time.
I look forward to seeing your tests and results!