I'm largely basing it on my own observation of both methods (I've used both
back and forth, and am currently (obviously) in the n-tier wins side).
I've read in a number of blogs and blog comments that it's a hit anywhere
from 10-20%.  Assuming that's true, it's highly situational, based on how
large the data set is.   I guess my warning should be worded better:

Be aware that using DataBinder.Eval has a performance penalty proportional
to the way you are using (width (columns) and size (rows) of the data).
However, it provides greater layer seperation and makes your code easier to
maintain and less brittle.  You should pick the best method for your given
scenerio.  Personally, I tend to value maintainability far more than
performance and will use the more maintainable approach first, profile, and
only adjust for performance when deemed necessary.

I like to make sure people understand both sides of a coin and make the best
choice for their situation....I think you and I do a decent job on that on
here :)

Cheers,
Karl