apparently this is well-known, but it wasn't to me: the A* convergence is very sensitive to the heuristic function.
if I write a heuristic that weights the top 2 rows more heavily than other rows, it converges much more quickly, but the path is generally much longer.
I found the diagonal H(x) function shown here to converge much more quickly than the Manhattan distance, for the 15-square puzzle.
even with the heuristic function that encourages speedier convergence, there is wide variance in the run time. Sometimes it finds the path in 10 seconds. Sometimes 10 minutes. Sometimes longer.
The number of moves required in the paths found, using the diagonal heuristic, ranges from 30-ish to 110.