Nyquist theorem is great, but, notice that decent oscilloscopes sample much higher than 2x the bandwidth. 10x is a good place to start to avoid any scaling and processing needs. Fancy scopes with complex sampling algo's will drop to as low as 3-5x the bandwidth. It also makes things like anti-aliasing much easier too.
Here's a nice white paper from agilent justifying going as low as 2.5x the bandwidth, but not all the way down to 2x (http://cp.literature.agilent.com/litweb/pdf/5990-9923EN.pdf