1) Downlink delay is a number of TTIs (subframes) by which the calculated feedback information from the base station (CQI,RI, PMI) is delayed before being employed at the UE side. It means the feedback information is outdated by this number of TTIs.
2) On a time-variant channel, the throughput decreases with increasing downlink delay because the channel changes between the feedback calculation from an estimated channel between the transmission for which the obtained feedback parameters are employed. So when a delay is introduced in the downlink, the link adaptation does not work perfectly anymore, because the channel changes with time.
The UE cannot compensate for the feedback delay (in our simulator). What can be done as compensation is channel extrapolation (prediction) before feedback calculation at the base station before calculating and signalling the feedback information to the UE.