I suspect that the problem is that when adding the additional coupling, you have added a very high frequency unstable pole.  Unstable poles are one of the main reasons why Spectre will stop due to `time step too small', and Spectre's careful timestep control makes it more likely that Spectre will nurture unstable modes. Furthermore, any round-off error when describing the inductance matrix of a large coupled inductor will cause it to lose its passivity.

I suggest that you set the spectre option `diagnose' to `yes' and rerun the simulation and note any messages that would indicate the solution is growing without bound.  If you have matlab, you could enter your inductance matrix and make sure it is positive definite.  If it is not, it will cause the circuit to be unstable.  Finally, you should consider increasing the resolution with which you specify the inductor values and the coupling factors.

The test for passivity of a two-terminal LTI inductor is that its inductance is positive [desoer and kuh].  The passivity criteria for a nonlinear two-terminal current-controlled inductor is that L(i) > 0 and dL/di > 0 for all i [desoer and kuh].  The passivity criteria for a four-terminal LTI inductor is that |M| < 1 [desoer and kuh]. Finally, the criteria for a 2n-terminal LTI current-controlled inductor is that L is symmetric and positive semi-definite [chua class notes].  For a flux-controlled inductor, the reciprocal inductance matrix must be symmetric and PSD.