Your data files indicate runs for 2005 and 2012. The 2012 is heavily oversaturated (queues of 700 veh+) and is grade-separated. The flow have been scaled for 2012 (from the base year of 2005). The CIRFLOW values (i.e. the peak 30-minute circulating flow) have been automatically calculated – here lies the problem! The initial automatic calculation of these values does NOT take account of suppressed demand and so the values used in the 2012 case are effectively wrong. The values for this case needs to be manually calculated as described in the Appendix of the Application Guide, taking account of the suppressed demand. The values for 2005 are probably OK. This will remove the effect described – i.e. increased delay in the 2012 case.
While on the subject of the effect of increasing demand flows, it is worth pointing out that increasing flows only on certain entries can have the effect of improving capacity on one or more entries. Let us take, for example, a simple three-arm roundabout where you are interested in the capacity of Arm A. An increase of entry traffic on Arm C will tend to increase the capacity on Arm A simply because traffic passing Arm B will be increased, which in turn, will reduce the capacity available to traffic entering on Arm B. This can reduce the circulating flow past Arm A hence result in an improvement in capacity on this arm. To summarise, if the changes in demand flow results in lowering the circulating flow past a particular arm, then the capacity for that arm will increase.