Thanks Johan,
OK so the two polys were in the _same_ database, which matches the issue confirmed in our testing.
The team responsible has been notified and they'll investigate the source of the performance issue. Thanks for letting us know!
FWIW the draw and data access of the single feature class without a join was confirmed by our regression tests and the team's testing to be much faster than 1.1.x, so clearly this is a case that needs to be looked into further for optimization.