Luke Wroblewski has been doing some research for his new book, Web Form Design Best Practices, through actual usage data. To that end, he has been working with Etre on several eye-tracking and usability studies focused on specific aspects of Web form design.

You can read more about here. What I did find interesting about the research was that form B (shown below) had the least fixations and the best visual distinction, while still giving equal weight to the primary and secondary actions.

Form B

As expected Form E did the worst of the lot.

“…Only Option E performed poorly during our testing. Six people mistakenly clicked on the “Cancel” button when attempting the task with this design, while many more lingered over it before realizing that they were about to make a mistake. People, as a whole, were around six seconds slower when using this design than they were when using Option B (a considerable gap when you consider that the placement of the buttons was the only thing that differed between the two). They also required a higher than average number of fixations to complete the task (with a higher than average total fixation length and average fixation length)…”

Form F

Which pretty much tells me that we should be forms the way we always have.