Single vs Multi-Pass Adaptive

I think it depends on the model geometry. If you're trying to capture stress in a small detail of a large component, you need to take more care than if the stress is on the outside of a simple shape. Looking closely at the shape of the contours will give an idea of how well it's converged (turn the plotting grid up).

If you really need accurate results, I wouldn't simply trust either SPA or MPA. The big advantage of MPA is that you can graph the convergence; I've had the most success using Mesh Controls until the convergence graph shows a good levelling-out, but once you've got the mesh right, SPA seems to give pretty similar numbers.

You should find them to be quite close, 10% or better, for most all applicaitons (often times <5%). Of course, there will be those that will vary by more than 10% or 15%, but for most of of what you'll likely find you need to do, SPA is the way to go The SPA solver is much faster as well, so it's great for BMX or Mechanica optimization studies. I would suggest using SPA while developing your components and then try MPA when you're complete to have it converge on the measure you're most interested in to see how well it works out. I've been using SPA for many years and have found it to work very well. I apply MPA in the manner I've suggested for about 25% of what I do - when it's a different enough part that has not been run before.