What is going on here?
 

This is a simple scalarTransportFoam test.

When I reduce the timestep, the scalar "freezes" prematurely. The smaller the timestep, the earlier the "freeze". See the attached pictures. Could I be running into machine epsilon problems?

This is a .1 meter square cavity with uniform flow from the left at 1.45e-8 m/s. The scalar is fixed value at 70 on the left side, so the attached graphs should level off around there.

The mesh is very simple - a 4x4 2-d grid.

Any insights as to what's going on?