The opinion here is that the question that will have the greatest impact on the practice of multiphase multicontaminant modeling is the extent of suitability of the LEA. If component partitioning is sufficiently rapid that it can be considered instantaneous on transport time scales, then the approaches of compositional petroleum reservoir simulation, appropriately modified and enhanced for groundwater applications, should be very promising. If not, the formulation of models may be quite different, and the computational burdens may be such as to require new approaches. This question heavily influences the level of detail of problems that can be solved within available computing resources.
Another important issue relates to fundamental understanding of multiphase flow: the form of constitutive relations, and in particular the manner in which they can be upscaled to the order of simulation grid sizes, especially in heterogeneous media. This ties in with the LEA question, because the important matter for modeling is whether LEA can be invoked at the grid scale. When state variables change substantially on smaller scales, use of LEA in averaged equations at grid scale may be invalid even when LEA is a reasonable approximation at the smaller scales.
These issues set the context for the question of choice of primary variables and how they are coupled in a multicomponent model. Numerical concerns include calculation of phase equilibria (needed whether LEA is invoked or not); types of spatial and temporal discretization, in particular the extent to which unknowns are treated implicitly or explicitly in time; linearization of nonlinear discrete equations; solution of linear equations; and efficiency in a variety of computing environments (scalar, vector, parallel).