I'm making a few assumptions here....but I'm pretty sure radar processors are accurate to within a pixel as displayed on the screen for range. However, they can only display the slant range. At its most extreme, an aircraft at FL360 could be over the radar head but displayed at a range of 6nm laterally from the radar head. This supposes that the radar is "visible" to the radar within the overhead. In reality its not really an issue, as aircraft which are close enough to each other to need close attention will be within the same portion of the radar's envelope.
For accuracy within azimuth, assuming the radar is correctly set up, the issue becomes one of PRF - pulse repetition frequency. For a radar head which rotates at 15rpm (5400 degrees a minute) and sends out 3600 pulses every minute, the distance in azimuth between pulses is 1.5 degrees. On the few remaining radar displays in the western world which can display raw, unprocessed primary returns, it's actually quite obvious. The blip shows up as an arc, and the longer the range from the radar head, the broader the arc. I've seen blips with a scale width of 2nm or more! All the radar can tell you is that the target is somewhere within that arc of 1.5 degrees. A processed radar doesn't have any greater accuracy - it just displays its best guess.
Whether that means the radar can conclusively show when an aircraft is busting airspace depends on several factors, including how the airspace boundary is alligned with the radar head. In the case of an arc around the radar head for example, the aircraft would show up slightly BEYOND the arc when actually directly overhead its boundary, due to slant range. How much beyond it would show would then differ with altitude.
So, the answer to teh question is.....it depends!
