Not me derailing the thread to sampling theory, ahem...
Anyway, bandwidth is an important concept for any kind of high-frequency instruments. The sampling theorem was mostly for the benefit of @py-bb asking if 1Gs/s 2 channel scope shouldn't be 500MHz, as well as to pre-empt anyone needing to point out to me that you can't turn 200Ms/s per channel into a full-fledged 200MHz scope, nor should anyone expect that sort of performance. (The sampling theorem doesn't fully hold, however, if the signal isn't fully unknown, which is why I've successfully used true 100MHz scope to align 200MHz bus signals, for example).
But the Nyquist-Shannon sampling theorem is a purely mathematical construct, and as such it assumes a perfectly spherical cow. In other words, the Shannon theorem paraphrased above actually goes "If a function x(t) contains no frequencies higher than B hertz, it is completely determined by giving its ordinates at a series of points spaced 1/(2B) seconds apart." Of course, the interesting signals in real world aren't pure sine waves (In fact, a perfect square wave is an infinite series of harmonics, to well, infinite frequency. Luckily you don't get a perfect square wave in practice, either) and hence 5 samples per frequency cycle is good minimum to interpolate min, max and zero crossings. And you need to also be able to get your 2 or 5 samples, where every component along the way has a bandwidth limit/attenuation.
I actually acknowledged that with "the analog paths likely can't". In retrospect, it makes sense that there would be an intentional filter around the actual capability of the scope's ADC's, so you don't get aliasing and other unwanted effects from higher frequencies. As a counterpoint though you could alter that filter, and for the signal-aligning use-cases I mentioned, you don't even care about significant attenuation as long as the peaks are discernible. But in the end, I don't expect the hardware can be pushed much further that way. It's just a thought because there ARE use-cases where you don't care about the exact waveform, or where the signal is repetitive so you could fold over the measurements (Unless the sampling rate is a perfect multiple of the signal, you don't even need to shift the sampling points, just align them in software).
Regarding modifying the FPGA, that's a thought, but reviewing some of the earlier thread it seems there's no way to re-program the FPGA via USB or SD-card for example (Could there be a hidden command for it though?) so while exciting, the actual number of people who would be doing that is probably fairly low... At least here in Europe the FNIRSI 1014D is less than half of the Hantek DSO2D15 for example, which places it a price category where you can buy one to toss around, experiment & possibly break, OR use if for hobbyist projects in the 20-30MHz range. But it's certainly cheaper to buy DSO2D15 or even 1013D than trying to replace the existing firmware for personal use. It does look like there's already something bootable for 1014D in the repo though, so who knows.