Very interesting post and also very interesting discussion, thanks.
We've done a lot of observation of low skills/low confidence users recently, so here are some extra thoughts based on their experiences.
Some people do not have email addresses or do not have access to their email.
Some low skills/low confidence users do not see any relevance to them of email, or feel unable to use it even if they had it. They may have limited access to computers - for example, using a friend or family member's computer when they do have access, or only using a computer in a library or some other public arena. They may have an email address in theory but not recall it, or they may not have an email address.
A signin process that requires email excludes these users. That may be fine for many sites, but it isn't acceptable for others - such as a government service where there is no alternative.
Some people have shared email addresses.
Another characteristic of low skills/low confidence users is that they may live in households where lack of money or lack of confidence means that they only have a single computer, and lack the skills to set up email accounts for each member of the household. The people in the household share an email account, and manage their email correspondence as a joint activity.
A signing process that requires a unique email address for each user excludes the second and subsequent user in these circumstances.
Label the buttons with what they do
You've placed the two options ("sign up" / "sign in") as two buttons close together in most of the suggested layouts. We've seen this work well as users see the two options and consider them as a pair.
We've also seen low skills/low confidence users get stuck as they don't understand the term 'sign in'. In one test, users correctly chose the other button (which had a site-specific name) solely because they didn't understand 'sign in' so opted for the other button as the least-worst available choice. When a user has low confidence to start with, any 'least worst' choice like this that further undermines confidence is a definite risk for breaking off.
So: having a pair of buttons to consider together: excellent. But careful choice of the text on the buttons is essential - and of course, testing that choice with your specific users.
Frequency of use of a term is not the same as it being understood in context
I agree with your analysis that 'sign in' is used more frequently than 'log in'.
Any high skills/high confidence user is very likely to know 'sign in', 'log in' and 'login' and what their effects are. If that's your audience, it won't matter much which you use.
But, as mentioned above, that doesn't mean that low skills/low confidence users will understand what you mean by it when they encounter it. So whichever one you choose, you may need to explain it.
Avoid two-column layouts
I don't think it's yet been mentioned in the column or discussion but it's worth repeating: don't use two-column layouts. I suspect you didn't bother to mention the issue because you know they're troublesome and wouldn't use them.
Don't put textbox labels inside the boxes
Another point that's worth mentioning explicitly: labels inside boxes are a disaster for low skills/low confidence users and for accessibility - and they're not great for many other users. I suspect this one is also so obvious as an issue that you thought it wasn't worth mentioning, but I see violations of this basic rule so often on sign-in forms that I think it needs to be stated explicitly.
Thanks again for a great article and great comments.