DNA replications needs a source of energy to proceed, this energy is gained by cleaving the 5'-triphosphate of the nucleotide that is added to the existing DNA chain. Any alternative polymerase mechanism needs to account for the source of the energy required for adding a nucleotide.
The simplest way one can imagine to perform reverse 3'-5' polymerization would be to use nucleotide-3'-triphosphate instead of the nucleotide-5'-triphosphate every existing polymerase uses. This would allow for a practically identical mechanism as existing polymerases, just with different nucleotides as substrates. The problem with this model is that ribonucleotide-3'-triphosphates are less stable under acidic conditions due to the neighbouring 2'-OH (though this obviously only applies for RNA, not for DNA).
So any 3'-5' polymerase would likely need to use the same nucleotide-5'-triphosphates as the 5'-3' polymerase. This would mean that the triphosphate providing the energy for addition of a new nucleotide would be on the DNA strand that is extended, and not on the newly added nucleotide.
One disadvantage of this approach is that nucleotide triphosphates spontaneously hydrolyze under aqeuous conditions. This is no significant problem for the 5'-3' polymerase, as the triphosphate is on the new nucleotide and the polymerase just has to find a new nucleotide. For the 3'-5' polymerase spontaneous hydrolysis is a problem because the triphosphate is on the growing chain. If that one gets hydrolyzed, the whole polymerization needs to be either aborted or the triphosphate need to be readded by some mechanism.
You can take a look at the article "A Model for the Evolution of Nucleotide Polymerase Directionality" by Joshua Ballanco and, Marc L. Mansfield for more information about this. They created a model on early polymerase evolution, though they don't reach any final conclusion.