This is still true (and is just as true of Ethereum, whether the devs on that project think of it as true or not). There is at least one instance in the past where the network has forked based on the (unspecified) behavior of bitcoind, and a couple more close calls where such forking behavior was properly reported and fixed before triggering.
If there is any way in which different implementations of consensus-critical code behave differently for the same inputs, it can be used to split the network. If you are running a node implementation other than the majority hash-rate reference client, you open yourself up to be potentially vulnerable during the fork. This remains just as true, if not more so with staking instead of proof-of-work.
You can do things to protect yourself like run ALL implementations and shut down if a fork is detected, although properly setting that up is nontrivial and in the end what value is gained? It very, very, VERY rarely makes sense to have multiple reference implementations of consensus code.
If there is any way in which different implementations of consensus-critical code behave differently for the same inputs, it can be used to split the network. If you are running a node implementation other than the majority hash-rate reference client, you open yourself up to be potentially vulnerable during the fork. This remains just as true, if not more so with staking instead of proof-of-work.
You can do things to protect yourself like run ALL implementations and shut down if a fork is detected, although properly setting that up is nontrivial and in the end what value is gained? It very, very, VERY rarely makes sense to have multiple reference implementations of consensus code.