The solution to this is simple, and it is that the blockchain (whose tip contains the most cumulative work) that follows all of the consensus rules is the Bitcoin blockchain, and any fork of this is not (in many cases, it would be an altcoin intentionally created).
The complexity comes in ensuring that the network is no longer partitioned and that everyone has received the blockchain with the most cumulative work.
The incentive to attack an implementation that is used by 10%-20% of the Bitcoin network is much smaller than the incentive to attack an implementation that affects 90% of the network. The further would be a minor hiccup, while the later has the potential to actually steal large amounts of money, and cause serious disruptions.
I disagree.
Suppose there is an exchange that happens to be connected to some nodes that are vulnerable to some kind of attack. Or perhaps they aren't connected directly to those nodes, but connected to node which are connected to those nodes. Suppose this attack causes those nodes to go offline or otherwise become disconnected from the network. Even if there are a very small number of these vulnerable nodes, if they happen to form a ring around the exchange, an attacker can attack those nodes and cause the network to partition. It would break into at least two pieces: the chunk containing the exchange, and the rest of the network. The attacker, if he has some hashrate, can now be mining a fork of the blockchain specifically created so that he can attack this exchange. Since this is a fork for a part of the network that is no longer receiving the rest of the blockchain, this attacker has 100% of the hashrate for that fork and can do everything with it that anyone with >51% of the hashrate can do. This kind of attack does not need a large number of nodes to be vulnerable, it just needs enough so that an attacker can partition the network.
Non vorrei dire fesserie ma, in questo caso non sarebbe invece piu semplice per gli exchange evitare queati tipi di attacchi visto che potrebbero avere magari 2 o 3 versioni diverse controllando che coincidano e bloccando tutto nel caso comincino ad esserci delle discrepanze? Se ci fossero piu versioni ed un exchange o chiunque altro si affidasse ad una sola versione allora tanto vale continuare con una versione unica come si fa ora.