While the other answers provide good explanation of what's actually happening, I think that they both miss the point of having 2 GSa/s on a 100 MHz scope.
The main point of interest is the way scopes generally do sampling. They often have a number of analog to digital converters which can be connected to different channels. The process which they often use to sample the signals is called interleaving. Basically the converters are set so that first one converter takes a sample of signal on a channel and starts processing it, then the next converter takes sample of the signal and starts processing it, then the third and so on until all converters took a sample. After that, first converter takes a sample again and the second and so on. So basically the cycle repeats. This allows use of slower and cheaper analog to digital converters, but has a negative effect on accuracy, since samples won't be perfectly equidistant.
So what happens when you have a two channel scope and use only one channel? Well all converters work only with that one channel and will provide the best representation of signal they can. But if you activate the second channel too, half of the converters will switch to the second channel and half will remain working with the first channel.
As it's already written, the rule of thumb is to have 1 GSa/s per 100 MHz of bandwidth. So if you take the 100 MHz scope which has sampling rate of 1 GSa/s, then you can effectively use only one channel at full bandwidth! If you want to use both channels, you can't use them with frequencies higher than 50 MHz, or you'll get sampling artifacts.
On the other hand, if you have a 2 GSa/s 100 MHz two channel scope, you can get better view of one 100 MHz signal or you can get a good view of two 100 MHz channels, which would be problematic with just 1 GSa/s scope.
So how does this apply to you: Well, let's take a look at product websites. For Rigol DS1102CA, it says under specifications Real-time Sample Rate 2 GSa/s（each channel），1 GSa/s（dual channels）, which means that the situation I explained applies here. Of the site for Rigol DS1102E, it says under specifications: Real-time Sample Rate 1 GSa/s（each channel），500 MSa/s（dual channels）.
So in the end DS1102E can work as 100 MHz one channel scope or a 50 MHz two channel scope, while Rigol DS1102CA is a real 100 MHz two channel scope.
A little bit extra info: As I previously said, it's bad for scope to use multiple analog to digital converters for a single channel, because the distance in time between samples won't exactly the same. This problem is initially solved by taking extreme care in routing clock signals for the converters so that the clock reaches all converters at the same time. Another (sometimes better) solution is to use multichannel converters. Usually it's easier to route the clock signal so that it reaches all channels on a single chip at the same time than it is to route the clock signal so that it reaches all physically separate chips at the same time. Some converters use other tricks too. For example one channel may be triggered at positive slope of the clock while the second may be triggered at the negative slope of the clock.
