spap2 returns the B-form of the spline, and if you don't supply a knot vector it will use C2 continuity (k-2 = 4-2 in your example) in the breaks.
The coefficients you were expecting are those of the pp-form of the spline, the coefficients of the power basis of the polynomials which form the spline. But because of the continuity conditions the 12 coefficients are not independent, there are 3 * 2 constraints, which give you only 6 independent coefficients (in the B- basis).
You can do fn2fm(spap2(3, 4, x, y), 'pp') to see the (3,4)-array of coefficients you were expecting.
Regarding the information on splines, the documentation in the Splines part of Curve Fitting Toolbox is a good companion to PGS. The notes in this link are useful, but the summations start in 0, not in 1, and that is a bit confusing (for me).