It sadly also depends a lot on the solver used and the way the problem was encoded in SMT. For a class in college I once tried to solve Fillomino puzzles using SMT. I programmed two solutions, one used a SAT encoding of Warshall's algorithm and another constructed spanning trees. One some puzzles the first solver required multiple hours whereas the second only needed a few seconds, while on other puzzles it was the complete opposite. My second encoding needed on hours for a puzzle which I could solve by hand in literally a few seconds. SAT and SMT solvers are extremely cool, but way incredibly unpredictable.