I've actually solved this, but I'm posting it for posterity.
I ran into a very odd issue with the DataGridView on my dual-monitor system. The issue manifests itself as an EXTREMELY slow repaint of the control (like 30 seconds for a full repaint), but only when it is on one of my screens. When on the other, the repaint speed is fine. I have an nVidia 8800 GT with the latest non-beta drivers (175. something). Is it a driver bug? I'll leave that up in the air, since I have to live with this particular configuration. (It does not happen on ATI cards, though...)
The paint speed has nothing to do with the cell contents, and custom drawing doesn't improve the performance at all - even when just painting a solid rectangle.
I later find out that placing a ElementHost (from the System.Windows.Forms.Integration namespace) on the form corrects the problem. It doesn't have to be messed with, it just needs to be a child of the form the DataGridView is also on. It can be resized to (0, 0) as long as the Visible property is true.
Not wanting to explicitly add the .Net 3/3.5 dependency to my app, I make a method to create this control at runtime (if it can) using reflection. It works, and at least it fails gracefully on machines that don't have the required library - it just goes back to being slow.
This method also lets me apply to fix while the app is running, making it easier to see what the WPF libraries are changing on my form (using Spy++).
After a lot of trial and error, I notice that enabling double buffering on the control itself (as opposed to just the form) corrects the issue!
So, you just need to make a custom class based off of DataGridView so you can enable its DoubleBuffering. That's it!
class CustomDataGridView: DataGridView
{
public CustomDataGridView()
{
DoubleBuffered = true;
}
}
As long as all of my instances of the grid are using this custom version, all is well. If I ever run into a situation caused by this where I'm not able to use the subclass solution (if I don't have the code), I suppose I could try to inject that control onto the form :) (although I'll be more likely to try using reflection to force the DoubleBuffered property on from the outside to once again avoid the dependency).
It is sad that such a trivially simple thing ate up so much of my time...