My first question is with regards to the size relationship between rlist and rvdw/rcoulomb, and consequently the meaning of rlist.
The option you want is rcoulomb < rlist and rvdw < rlist. This works and gives the most accurate and also the most costly integration.
A faster and very commonly used setting is: rlist=rcoulomb=rvdw. With PME the Coulomb interactions are very small at the cut-off, as are the LJ interactions. So with a small sacrifice in integration accuracy one can gain a lot of performance, also because analytical LJ is cheaper than tabulated.
The last option is rcoulomb > rlist and/or rvdw > rlist. Then the energies and forces beyond rlist are only updated every nslist steps. This gives less integration accuracy but can give a lot of “interaction” accuracy at a small computational cost.