The behavior you're describing sounds like a job for UITableViewCellAccessoryDetailDisclosureButton, actually. If you're going to be pushing another view onto the stack but don't want to do that when the user selects the cell proper, use a detail disclosure button to maximize your consistency with existing UI convention. It's hard to say without more information.
As for doing the work necessary to detect touches in a sub-region, that may be more trouble than it's worth. True, adding subviews to a cell incurs a compositing cost but if you're only talking about one button, and a button whose background you can set to opaque, you'll probably be just fine. The alternative is reinventing the wheel to recreate button behavior in a subregion of the cell and that doesn't sound like functionality that will be much fun to maintain as the SDK matures.
That said, adding views to a cell doesn't incur a compositing cost per se, it's drawing those views that's the trouble. So if you really wanted to go nuts on the optimization, you could create a pre-rendered cell background image that includes the appearance of a button you want and then place a custom, image-free, see-through UIButton instance right over top of it. Nothing to draw, so no additional compositing cost. Worth a shot.
You didn't come here for a premature optimization speech, so I won't bother with one, but I say just do the cell with a normal button for now, make sure you like how the functionality feels to use then optimize toward the end if you're looking at performance that you're not happy with.