LES combustion problem
 

I am modelling non-premixed turbulent flames using LES. I am systematically obtaining a much smaller jet penetration than experiments or RANS solutions. I even obtained it starting from a "correct" k-e solution. I have tried several combustion models and the problems persist. (it also happens when I test a pure mixing case with a prescribed density distribution in mixture fraction space) I think is related to the relatively coarse grid used that introduced a lot of numerical diffusion. However, some people suggest me that incorrect incoming turbulent BC can be the cause. Any ideas/suggestions around there?