Just to define the term for everyone, Equivalent Time Sampling is when you take advantage of the fact that a waveform repeats, so by sampling 10 different cycles of the same waveform at 50kHz with slightly different offsets on the ADC timing, you can recover a signal sampled at 500kHz. This only works if every cycle of the signal is identical, though.
The way I understand EST, and correct me if I'm wrong, is that you have lower-rate ADCs sampling at a 360/n degree offsets. So you might have a 100 kHz clock driving two 50 kHz ADCs at offsets, giving every second clock pulse to each respective ADC. In other words, you
are sampling from the same waveform.
But unless they've overclocked the ADCs (or come up with a way to disprove the Nyquist-Shannon theorem :p ) that won't get them anywhere near the claimed 250 kHz bandwidth. Actually, looking closer at the datasheet, what is claimed is that the
full power signal bandwidth of the ADC is 50 kHz, which I guess means there's a roll-off above that frequency? The conversion time of the ADC is 5 µs, giving an effective sample rate of 200 kHz if run at the maximum speed, or for two running in parallel, 400 kHz, giving a crusty 200 kHz effective Nyquist bandwidth. Not quite 250 kHz, but we're starting to get into the plausible territory.