I beleive that set-up time for a repeated START condition in i2c_slave_model.v is wrong. Line 340:
fast_tsu_sta = 1300,
should be 600 according to the Phillips specification. This will fix the tsu violation error in Modelsim.
So tsu_sta is 600 (min). And the core uses 1300, because that's how it fits in the statemachine. How is that a bug?!