[en] In fission gas release modelling, it is normally assumed that any locally released gas mixes instantly and perfectly with other gases throughout the internal rod void volume. The present work investigates the consequences of the assumption that perfect mixing is dependent on diffusion to occur, although the sub-assumption is maintained that pressure equilibrium is instantly achieved. In other words, when a burst of gas release occurs at any axial location, sufficient local accomodation takes place throughout the rod to eliminate any pressure gradients, but due to the narrowness of the passages through fuel cracks and fuel-cladding gap, concentration gradients may still prevail. Diffusion coefficients for the subsequent concentration equilibration are derived from classical theories. Application of one-dimensional diffusion theory is straightforward, but the lack of knowledge of the effective width of the axial passage introduces an uncertainty. If the assumption is made that the total cross-sectional void-area at any cross-section is available for diffusion through a straight, cylindrical duct, the theoretical exercise shows that the diffusional delay is unimportant, since equilibration will then happen at a faster rate than gas release. If, however, one takes account of the fact that the diffusion path is distorted due to the random positions of and surface mismatches between fuel fragments, the effective duct area is very much reduced, and the diffusional delay may turn out to be of great significance. This path distortion is modelled by multiplying the cross-sectional void area by a ''constriction factor''. The model has been implemented in a developmental performance code and benchmarked against power reactor and test reactor data. By carefully adjusting the constriction factor, it was possible to considerably improve the overall fit of the predicted gas release to the benchmarking data base. However, only rods with a long fuel stack and/or sufficiently peaked power histories were sensitive to the diffusional delays. Typically, short test rods were not affected. (author)