it costs about 500ns to start a memory to memory EDMA, is it normal?

I am trying to speed up copying data from inner memory to SDRAM, the DSP works at 400MHz. It seem it costs >500ns to set and start the EDMA, for example following code costs 134 CPU clock, it's about 340ns:

edma3ccRegs->PARAMSET[10].DST = DST;  
edma3ccRegs->ESR |= 1 << 10;   

However, write 16bytes to SDRAM by CPU only costs 42 CPU clock.

it seems that there is no benefit to use EDMA to copy the data if the data size is small.

In my applicaiton I need to copy about 64 bytes from inner memory to SDRAM every 5us. I want to do this as fast as possible.