may be this discussion can help you. http://www.cypress.com/?app=forum&id=2232&rID=93505
Similar results I have found on 050B kit at 115200 Baudrate with SwTxUART with Instruction Cache Disabled
Thanks Viktor. I think I read that thread when I first started working with the UART. At this point I'm considering configuring some form of an external clock source to the get to 921,600. It makes sense that what I'm seeing shows up in the upper nibble as the accumulated tolerance stacks up across a byte.