Huh. I thought I remember Kiran giving a really good answer to this question at tea in Berkeley, involving K3 surfaces.
Well, to make up for that, I'll point to some references which definitely contain answers. Look at section 2.3 of Ravi Vakil's paper on Murphy's law for deformation spaces http://front.math.ucdavis.edu/0411.5469 for some history, and several good references. Moreover, Ravi describes how to build an explicit cover of P^2 in characteristic p which does not deform to characteristic 0. Basically, the idea is to take a collection of lines in P^2 which doesn't deform to characteristic 0 and take a branched cover over those lines. For example, you could take that p^2+p+1 lines that have coefficients in F_p.
PS According to Ravi's paper, you are wrong when you say that this example can't be projective.