There is latency involved in both production and detection to contend with.

NYC to Tokyo would require an unprecedented beamline angled deep down into the earth as well as a much more powerful proton driver to produce enough neutrinos to compensate for the 1/r^2 loss of flux.

And, then there is the neutrino oscillation resonance due to the high matter density as the neutrinos pass near the Earth's core just to mix things up.

The price tag for DUNE is a couple billion USD for a relatively "short" baseline of 1300 km and 40kton far detector.

All that said, if some fat cat trader wants to fund it, I'll help build it.

Any individual bit could be noise, so it might take 2-3 bits to transmit something unambiguous (via an error-correcting code or similar).

If so, I think the speed advantage disappears.

Detecting one pulse above background noise is not terrible, according to the paper. However, the way the paper sets up their communications channel is kind of silly, since they are pulse rate limited.

If you wanted to transmit a lot of information with one pulse, you'd use time domain slicing and synchronized atomic clocks.

Regardless, you don't need particularly high signal fidelity in order to arbitrage the market, just better than 50/50. The ~25ms they can cut off of the transmission latency is an eternity in market time.