...
What would happen in your example is the millions of bitcoin participants would tell that one miner to take a hike and continue with the existing P2P core protocol. Then when that miner started to produce nonconforming 25BTC blocks, they would simply be rejected out of the P2P network as invalid, and the honest network would carry on. The attacking miner could have 95% of hash power and it wouldn't matter, his blocks would be rejected as nonconforming and the network would simply re-adjust to the 5% remaining.
...
IIRC that actually did happen at the 50BTC->25BTC juncture. What you 'predict' is about exactly what happened I think but I was not paying that close of attention. Whatever happened it was a blip and over within hours or days.
Back then though more Joe Sixpacks were running transfer nodes...I was at the time but I've since stopped (in part because I'm behind a satellite connection and the blockchain represents several months of my data allowance.)
edit - Nowadays with most people running Multibit, most people may not know (or much care) what chain they might be following in a stolfi-type attack scenario. But then again they may...I don't know the precise details and capabilities of Multibit.