Tom, your statistical evaluation is very interesting, thank you for this!
But as you write, the deviation reported by SPM's autoalignment cannot be fully trusted. I don't really understand why, but my experience is that this number rarely corresponds to what I come up with by measuring manually, and many times it is absolutely off.
Back to the original question George posted:
I often bracket the base when taking close up shots with a beamsplitter macro-box. So I end up with a series of images with various deviations to choose from.
My personal experience is that close-ups of flowers and similar objects often look more natural with a much smaller deviation than 1/30 in projection. Many images I project have as little deviation as 1/60 or 1/80, even if I had a whole range of versions to choose from.
So if I were a judge, I would definitely not consider 1/72 or 1/42 to be a problem by itself. The resulting impression depends on the object and composition.