Not quite as simple as it seems.
"how big a deal it would be to have the ability to shed some load for about 15 minutes. The faster the load shedding can happen in response to a request (100s of msec would be awesome, seconds would be great, minutes would still be very helpful) the more valuable the ability is."
See also e.g. the UK's small number of grid-scale pumped storage stations such as Dinorwig. Oiriginally built decades ago to smooth out demand on the scale of a few hours, to allow the grid to operate sensibly with the inflexible output from nuclear generators, Dinorwig has a response time in secinds *and* high power connectivity (output of a GW or so, from cold, in a few seconds. Not a few tens of MW) which is now marketed to provide a "grid scale fast response frequency regulation support" resource.
Ref: https://en.wikipedia.org/wiki/Dinorwig_Power_Station
Dinorwig is no longer used as a grid-scale energy supplement for a few hours at a time. "leaving it to the market" meant that job got delegated to fields full of non-functioning grid-scale diesel generators a few years ago (insane, but potentially profitable).
Basic load shedding at a domestic level could be done quite simply and cheaply with the kind of "dynamic demand" management technology that's been around (but never deployed at scale) for years. See e.g.
https://www.dynamicdemand.co.uk/