Extremely high contrasts in conductivity such as the ones introduced by a multi-layer well in low-k material create some difficultiy for the solver. First of all, please check whether you are using SAMG, and - if yes - what the internal error criterion is. This should be small, e.g., 1e-12. Otherwise SAMG may terminate the iteration too quickly, producing somewhat random results. Then, make sure time steps are small during cone development. In case of time-dependent abstraction, smooth changes in rate may also help.
