As long as by arb we're in agreement that it is purely statistical in nature with a slight edge in knowing information on delivery points and individual dynamics.
Because unlike WTI/Brent where one could theoretically delivery one product in place of another this would be impossible with natural gas. Thus,
I would think any premium/discount that the futures trade to the spot between markets would be a product of their own localized supply and demand and unless these relationships got particularly skewed one could not make a high a probability trade.
In the case of natural gas in particular, the supply is controlled by turning a valve on or off and storage when and if demand declines is very easy for "producers." As such, I could see minor deviations occurring based on localized disturbances that might not "revert" and major disturbances as a result of natural disaster and other shocks that might be even more prolonged.
That being said, I am intrigued and I'd be interested in learning more about your strategy. What would be the time to be in a trade? Is the trade trending in nature or mean reverting? Have you tried this strategy with WTI/Brent? or is that, per your previous post, too efficient a market in your estimation?