This is making me a little crazy, so bear with me.
If you choose 12pt text in InDesign, and then export the document to ePub, InDesign creates CSS that calls for a font size of 1em.
When iBooks looks at that CSS, it translates the 1 em into 16 pixels. But since the iPad has a resolution of 132 pixels, the text's actual, physical size isn't the same as if it were on a monitor, where we're most used to seeing pixel sizes for fonts. Indeed, on the iPad, one pixel is 1/132", so 16 pixels are .12". And since in the physical world, 1" = 72 pts, that text that you originally sized at 12 pts, is now only about 9 physical points in iBooks (.12" x 72 = 8.64).
If you displayed the same XHTML document on my monitor (at about 98dpi), where 1 pixel is 1/98", that would mean 16 pixels would be about .16", or about 12 pts—the same physical size, if you put a measuring tape up to the monitor, as if you printed it on a piece of paper.
This all came to a head since I've been creating ePubs with Microsoft Word 2007 on Windows on Parallels on my same 98dpi monitor. Again, I choose to set text at 12 points. If I print the document on a piece of paper, it measures 12 actual points.
But when I save it as HTML and then convert it into an ePub, Word generates CSS that actually specifies 12 points.
So how does iBooks and the iPad in general deal with points specified in CSS? Curiously, it's very similar to what happens if you specify 12 points in InDesign. iBooks interprets 12 points in CSS as 16px, and so outputs it exactly the same as if you had specified 1em (or 16px) and so the text is actually displayed at about 9 physical points, not the 12 that you specified.
What's the moral of the story? How should you specify the size of text destined for an ePub to be read in iBooks (and elsewhere)? That's is an even more complex issue. Surprisingly, it turns out that the most variety of outputs in my tests was from using ems. ADE and Ibis Reader on the Mac (but not iPad) displayed 1 em at 21 and 19 pixels, respectively, while all the ereaders that I tested displayed 16 pixels and 12 pts as 16 actual pixels (and 9 actual points!)
(Line-heights are another issue altogether since there is a wide variety of default line-heights from one ereader to the next.)
So I'm tempted to recommend using pixels over ems. And I'm finally truly understanding, with my heart and not just my brain, why pixels are a relative and not an absolute unit of measurement.