For 3), it is a trade-off for losing the convenience from SUSY, but I guess under small N, i.e. finite dimensions of matrices, matrix model computations are straightforwardly linear algebra equations.

For 1), I don't know if there is a large gap in the regulated YM until the further calculation is done, but a gap, potentially enlarged by quantum corrections, is observed in SYM matrix model in 1401.2020. To consider a proper standard model, I agree SUSY is necessary but maybe not so crucial for the duality? For 2), stabilization is supported by the regulator, i.e. the Lagrange multipliers, in the matrix YM model, see 1510.05779.

Certainly not the gravity dual to pure YM. Due to the flat direction of YM and therefore divergent partition function, I think there is no dual gravity. However, the YM with extra terms such as Lagrange multipliers might be doable when YM is regulated. The calculations in my mind are like t'Hooft's old computations in the matrix model. These are much easier without pfaffian, even numerically calculable with Monte Carlo methods.