There were some papers and articles published about the deanonymization the bitcoin network and such, reading those may help you understanding how such techniques work. One of them e.g.
http://www.coindesk.com/eavesdropping-attack-can-unmask-60-bitcoin-clients/ (June 2014)
Informal description of the client deanonymization attack on the Bitcoin P2P network"The researchers also found that the attack could be designed to prevent the use of the Tor network, which anonymises traffic. Additionally, the attack can 'glue' transactions, so that transactions performed on one machine using multiple bitcoin addresses can be grouped together."(Also asking here, is this concept/theory correct?)
Something like this tells me that if all transactions (and hence addresses) in a wallet can be tied together, then if a single transaction reveals your identity (at any point in time), your wallet and subsequent transactions are compromised. Assuming someone is willing to invest the time and effort.
Yes, the concept/theory of a poisoned node is correct. It is because, like Skype actually, every time you use a full wallet (like Armory) it broadcasts your IP address every time you boot up and/or send or receive (it's not clear to me which, but it's one or the other). From this information a dedicated person can track your activities--if they have specialized software like the researchers from the paper below. Or, as in the paper involving Tor, they can set up 'poisoned nodes' next to yours (possibly geographically located, so they are the first nodes to greet your PC when you boot up), and intercept your bitcoin communications that you think are heading to Tor, but in fact are going to the poisoned nodes. But to do so, they need to have lots of such bad, poisoned nodes (the article mentions renting lots of servers), to trick the good nodes, which is something only either a big government agency or a very sophisticated criminal syndicate can afford to do, not your casual basement hacker.
TonyT
https://www.getbitcoin.com.au/bitcoin-news/investigation-white-paper-anonymous-bitcoin Although numerous Bitcoin clients exist, none of them are specialized for data collection. Available clients often need to balance receiving and spending bitcoins, vetting and rejecting invalid transactions, maintaining a user's wallet, mining bitcoins, and, perhaps most detrimental to our study, disconnecting from "poorly-behaving" peers; these were precisely the peers we were interested in. Because existing software had integrated functionality that interfered with our goals, we decided to build our own Bitcoin client called CoinSeer, which was a lean tool designed exclusively for data collection. For 5 months, between July 24, 2012 and January 2, 2013, CoinSeer created an outbound connection to every listening peer whose IP address was advertised on the Bitcoin network. We maintained that connection until either the remote peer hung up or timed out. In any given hour, we were connected to a median of 2,678 peers; for the duration of our collection period, we consistently maintained more connections than the only other Bitcoin superclient we know of - blockchain.info. This data collection effort required storing 60 GB of data per week