you can write your own driver which sends out as many bytes as it can at each interrupt instead of the SDK one which, unless it's changed, sends one byte only then waits for another interrupt before it does the next one.
no you should most definitely use the hardware TWI peripheral, but the driver, if, as I said, it's not changed, puts huge gaps between characters at higher speeds, so modify it to send more than one byte during the interrupt service routine if they're going out fast enough.
