The phase=1,2 is for phase information in the indirect (f1) dimension. For each 2D increment, two FIDs are recorded, one with the pulse sequence where phase=1, and another one where phase=2. The value of "phase" will affect a number of phases or for example PFG pulse(s), depending on the pulse sequence/experiment. This is not to be confused with the regular phase cycling, as each FID will have the usual phase cycling for each transient, and each FID consists of nt transients.

So you if you have 128 increments (ni=128) in your experiment, you'll end up with 256 FIDs. How the FIDs are used in 2D Fourier transform is determined by f1coef parameter in vnmr/vnmrj (basically sequence of multipliers for the 1D FT results, see varian manuals for details). The two FIDs basically correspond to real and imaginary parts for the f1 dimension, and are used to get sine/cosine modulated components for phase sensitive f1 / 2nd dimension. This is known as the States-Haberkorn-Ruben method:

http://www.chemie.uni-hamburg.de/nmr/insensitive/tutorial/en.lproj/purephasespectra.html

and more details (page 7-25): http://www-keeler.ch.cam.ac.uk/lectures/understanding/chapter_7.pdf

If I remember correctly, Varian sequences use tradditionally this method, while Bruker uses the TPPI method to achieve basically the same thing. I have no idea how to program the spectrometer to obtain two fids or then process the FIDs correctly in Bruker machines, as I have done pulse sequcence programming only in Varian. Of course you can also modify the phases to implement TPPI style aquisition.