I posted earlier about the difference between DataBinder.Eval and 'strongly typed' binding - mainly to see what other people do...Well in that I mentioned a benchmark I did a while ago to compare the two which showed a roughly 20% difference. So, in the interests of transparency, I've rewritten the benchmark - not looking to win any coding contests, but it should be fairly accurate!
Anyway, if you want to try it out / pick fault with my methods (there's probably a few problems, I just hacked it together) you can get it here.
It basically binds on to two DataGrids multiple times (20 times each in the version I uploaded so it runs 'fairly' quickly), one DataGrid uses DataBinder.Eval, one uses the 'strongly typed' method, it uses a "Select *" from the Orders table in the SQL Server 2000 Northwind database - 830 rows of fairly realistic data.
I use a dataset which I rebind on to multiple times; to get over the impact of the SQL server query, the data is loaded once only. I also run each type twice in 1:2:1:2 order to try to reduce the impact of order effects. I've also run the test in reverse order and with fewer and greater loops (up to 2000 for each grid). My results are fairly consistent at around 20%  (varies between around 15 - 30%) better performance for the 'strongly typed' method.
I hope someone downloads this / does their own test, I have no particular affiliation to either method,but I do  tend to use the strongly typed one, but performance is a secondary concern to the explicitness of that method.
Anyway, if you do try this out, please let me know your results.
I ran this test on my home dev machine, an Athlon XP 2600+ with 1GB PC3700 DDR RAM running .NET 1.1.4322, SQL Server 2000 Dev SP3, Windows XP (all latest patches applied).