Error control coding in the context of joint detection in multiple access channels is addressed. Special attention is given to iterative (turbo) processing proposed in the late 90's, which can be viewed as the serial concatenation of an error control code and the channel and decoded as such. It is shown that separating the coding from the multiple access demodulation is optimal for higher signal-to-noise ratios, and can be achieved with a simple iterative demodulator using spatial coupling with external coding. In the low signal-to-noise ratio regime, mean-square error filtering followed by error control coding is optimal. Full iterative integrating of the error control codes is never the best solution.