DMA Controller

I want the most efficient memory usage possible. So I'd like all the reads/writes to memory to use the full 32 bit datawidth. The problem is my peripheral has an 8bit databus. Initially I set the datawidth for the dma streaming to 8bit during testing and it worked fine. The main question was whether it would automatically buffer 4 bytes of those to maximize the memory efficiency. I don't want to waste memory cycles by only writing a byte during every cycle. I'd rather redesign the paripheral so that it would buffer 4 bytes and then do 32 bit transfers. Before I do that I wanted to make sure that it wasn't already being done automatically.