The problem with 2PC is that a malicious node could stop somebody from being able to spend their coins by always sending nays out to the network when ever the victim sent a transaction. To prevent this, you would need to be able to detect when a node is faulty/malicious which would require implementing a costly Byzantine Consensus Protocol [1]. In practical systems that can withstand Byzantine faults, the number of messages required to agree on a log entry (e.g. a transaction) would be O(n^2) in the number of nodes in the network, which would greatly limit scalability.
The genius of Bitcoin's Proof of Work protocol is that:
1) it is more resilient than Byzantine agreement. Byzantine agreement with 3f+1 nodes can handle at most f faulty nodes, while Bitcoin can tolerate 49% of the network hashing power being malicious. (though this is being debated currently due to theoretical strategies like selfish-mining)
2) it is much more efficient than Byzantine agreement in the number of messages sent, which has allowed the network to scale to thousands of nodes, although they have been running into issues with the blocksize/block limit, but their are currently research efforts [4] underway to remedy this.
The main problem with Bitcoin's proof of work scheme it is that it is extremely expensive in terms of CPU cycles, but this is solved by compensating miners for their efforts through coin generation and transaction fees.
[1] - http://www.cs.cornell.edu/courses/cs614/2004sp/papers/lsp82.... [2] - http://pmg.csail.mit.edu/papers/osdi99.pdf [3] - http://arxiv.org/pdf/1311.0243v5.pdf [4] - http://www.cs.huji.ac.il/~avivz/pubs/13/btc_scalability_full...