First, FFTs do not work well on small data sets, two cycles of the lower frequency in your example. I increased the number of samples in the function generators by a factor of 100 and got much sharper spectra.
Second, What phase are you trying to measure? What is the phase of a complicated signal, measured with respect to what reference? If your signals are not exactly harmonically related the phase of one of them with respect to the zero crossing of the other is meaninless as it changes from cycle to cycle?
The FFT is defined mathematically for a continuous signal extending to plus and minus infinity in time. The phase associated with the FFT is arctangent(Imaginary part/Real part).
Lynn