Everytime you fire a PKTEND, one buffer is wrapped. (Which means you can no longer fill data in that buffers. At the end of 8 PKTEND you will have 8 buffers none of which can accept any more data. (Though the buffer size is 16K, asserting the PKTEND will wrap each buffer, so the new data will always go to next buffer).
After 8 PKTEND, the 8 buffers are wrapped and no more data can enter into FX3's DMA from FPGA. So flag A will wait as long as the data is consiumed by the PC.
If your PC is receiving large amounts of data rapidly (just like how Fx3 C++ or C# streamer application does), then there will not be such long delay.
- Madhu Sudhan