I think the massive electromagnetic flux caused by the CME interacting with Earth's magnetosphere induces current even in powered-down systems. I believe it's necessary to cut lines and perhaps even disconnect individual transformers from the grid.

Absolutely, but transformers can already be disconnected from the grid, you just have to have the guts to do it in advance.

How long does that take to do though? My understanding is that the scale of a CME is what makes a potential disaster.

The major distribution sub stations already have remote controlled disconnects that isolate them from the power grid it's a basic requirement for maintenance.

They can have some pretty powerful arcs too: https://www.youtube.com/watch?v=A5YxdRfAe14

Interestingly the UK national grid's current policy is the opposite of this - i.e. to open up as many circuits as possible to provide multiple paths for charge to dissipate through.

The power grid can't be shut down in a few hours.

One of my college friends who works for PG&E said "The faster you turn off the grid the longer it will take to turn it on."

He had some great stories about contingency plans that go into how they would restart the grid post 9.0 quake in SF or LA. Such a quake is largely expected to bring down most, if not all, of California's grid.

The strike slip nature of the San Andreas Fault zone in California precludes really big earthquakes. USGS thinks a 8.2 or 8.3 is about as big as they could get. The biggest historical California earthquakes have reached 7.9 in magnitude. (http://www.conservation.ca.gov/index/Earthquakes/Pages/qh_ea...)

However Northern California could see a Mag 9.0+ quake from the cascadia fault.

The mean advance warning time would be ~3 days, it says.