CfB and mthcl, the onus is on you to prove and demonstrate what you have given up from the CAP theorem. You have not. Period.
Consistency is weakened to eventual.
To make sure we are on the same page: what is partition tolerance and availability of Bitcoin?
If Consistency is weakened to eventual, then either you have no defined Consistency (i.e. no Consistency ever) or you have an equation for probability of Consistency. If there exists such an equation, then you have to explain how and the probability of either Availability or Partition tolerance is lost when the probability of Consistency is attained. The onus is on your to justify these claims analytically, including convincing arguments about the game theory. Else you can just put it into the wild and observe (and who knows what will happen).
Consistency in Bitcoin is the fact that the objectivity is the longest chain. There only state of inconsistency is the probability of an orphaned chain, which declines over time except if the adversary has greater than 50% of the sustained network proof-of-work hashrate.
Availability in Bitcoin is given by even if there are no other active nodes, then sender and/or recipient of the transaction can extend the longest chain and the Consistency remains valid (except for the caveats on Consistency).
Partition tolerance is lost in Bitcoin because if there is network partitioning then double-spends can occur on each chain without being detected until these chains are merged. Bitcoin can't tolerate multiple chains, and only allows the longest chain. There is no way to merge these chains, because double-spends can infect other downstream transactions, combined with inputs from legitimate transaction graphs.
So what we can say is Bitcoin fulfills the CAP theorem, except it has theoretically unnecessary caveats in Consistency due to 51% attack and delay due to probability of orphaned chains, which are due to the ephemeral Partition tolerance that is introduced by chain reorganization. The Consistency delay also causes transaction confirmation to be significantly delayed. The goals of my Sync (or BlocSync) block chain overhauled design has been to eliminate those caveats, while relaxing the Consistency and/or Availability during partitioning of the network in order to provide some Partition tolerance.
If the CAP theorem is random to Byzantine fault tolerance for networks, then we might as well say the speed-of-light is an irrelevant fact when doing Physics.