You can use multiple systems in parallel timed so that they sample in turn. Each ADC can have its own dedicated memory and control but the timing is in concert. Here's an example:
Suppose that you had 2 ADC that each sample at 50 MHz but timed to sample in this order:
ADC1 - Samples a0
ADC2 - Samples a1 (10 ns later)
ADC1 - Samples a2 (10 ns later, but 20 ns after the prior ADC1 sample)
ADC2 - Samples a3 (10 ns later, but 20 ns after the prior ADC2 sample)
etc.
It would be easy to read the memory in such a way as to reconstruct the data (i.e. [a0, a1, a2, a3, ...]). The equivalent sampling would be 100 MHz.
In the case of a repetitive waveform, you can do the same thing with a single ADC sampling at various timings after multiple triggers. The assumption is that the waveform isn't changing so you can sample it differently over multiple periods and represent it as being sampled over one period. Here's an example:
Trigger Event
ADC - Samples a0 (10 ns after trigger)
ADC - Samples a2 (20 ns later, 30 ns after trigger)
...
Next Trigger Event
ADC - Samples a1 (20 ns after trigger)
ADC - Samples a3(20 ns later, 40 ns after trigger)
...
Reconstruct with order [a0, a1, a2, a3...] and the equivalent sampling rate is 100 MHz.
-jonathan