Charges from Huckel and extended Huckel theory rely only on the connectivity of the system. To the extent that Gasteiger charges (and its several variants) are elaborations of Huckel theory, I would expect these would give reliable charges, within the framework of that theory. In fact, adding Gasteiger-Huckel charges to a 2D molecule (MDL sdf files) before optimizing the molecule in Sybyl (using a force field, which captures the connectivity information) is a fairly common task in our lab. You can do these calculations in Alchemy (cheap) or VEGA (free!). You might also check TINKER (as a free/cheapware), but I am not experienced in its use.
Doing 1SCF in MOPAC on a 2D molecule will give you meaningless charges (not just inaccurate!). This is because MOPAC's (NDO) methods do not depend on the "bonding" information implied by a connection table - bonding is inferred after a stable (optimized) configuration of the nuclei in the electron soup is calculated. The charges will be calculated (you can always do the calculation!) but the 2D molecule is so far from a stable configuration that heaven alone knows what they will mean.
"Inflating" a 2D molecule to 3D in MOPAC _may_ work, but again, since bonding is inferred after optimization, there is a good chance that your structures will be ... ummm, novel? This is particularly true if you start with Cartesian coordinates. With the Z-matrix (internal coordinates) ... maybe yes, maybe not. I've had some wonderful experiences with poorly defined (in 3-space) starting structures giving very strange molecules after optimization when I drew in the lines representing what I thought were bonds. At best, you will need to carefully review each structure after the calculation.
Hope this helps,
Steve