The hypothetical solution is flawed: it would take not 50% of miners to adopt a 1% implementation; it would take 100%, since if only 1 defects, that one wins all the coins and all the cooperative 'prisoners' lose.
There are two mechanisms by which defectors could win coins.
First, they could simply mine more coins. For them to win almost all the (new) coins this way, they have to do almost all the mining. Say, 80% of it, although it's really a matter of what you mean by "almost all".
Second, they could collude and subvert transaction audit trails so that the Bitcoin system transfers ownership of all existing coins to them. For them to win all the (new and old) coins this way, they only have to do more than 50% of the mining.
So, if collusion is feasible, the second approach is easier; so let's forget about the first one for the moment.
The point at which the defectors, each using 2 orders of magnitude more CPU time, are doing more than 50% of the mining, is when they control more than about 0.9901% of the network's total raw computing power, because at that point, the cooperators control only 99.0099% of the network's total raw computing power, and the cooperators are using only 0.990099% of the network's total raw computing power to prevent counterfeiting.
If a single defector controls only, say, 0.5% of the network's computing power, they will get only about a third of the new bitcoins mined, and will not be able to subvert the network.
Only needing to have 0.99% of the network collude in defection to break the system is a pretty bad problem, but it's not nearly as bad as any single person being able to break the system.
