This is an experimental physicist's answer:
The linked article is careful to state:
That means that conservation of energy can appear to be violated, but only for small values of $t$ (time)
Italics mine.
Conservation of energy is an experimental fact that has been validated in innumerable experiments. This means, as far as the correspondence with quantum mechanical formulations goes, the energy is an observable and to every observable there corresponds a quantum mechanical operator, which, operating on the wave function of the system under observation, gives the value of the observable. Quantum mechanical operators either commute, anticommute or are non commuting and when they are not commuting a corresponding Heisenberg Uncertainty relation will be defined for the expectation values from the non-commuting operators. The Heisenberh Uncertainty Principle is a generic expression of the non-commutability
Reading further in the link we see:
This allows the creation of particle-antiparticle pairs of virtual particles.
virtual means "as if", and it is a mathematical construct which affects the calculations for cross sections and lifetimes and other measurable quantities but cannot be measured by itself. In these simple Feynman diagrams:
There are real particles and virtual particles, Real are the incoming and outgoing lines from the vertices, and virtual are the internal lines connecting the vertices. The real lines represent real particles on their mass shell. The virtual lines carry the quantum numbers of the particles with their name but are not on the mass shell, i.e. you cannot apply momentum and energy conservation at the same time because the mass corresponding to the virtual particle is arbitrary (well depends on probabilistic mathematical functions but arbitrary for the argument). Look at the diagram responsible for weak decays: the mass of the $W$ is $80.4\ \mathrm{GeV/c^2}$ and is the virtual particle exchanged in the beta decay of the neutron whose mass is less than $1\ \mathrm{Gev/c^2}$.
So when one reads "virtual particle creation" coupled with "apparent energy non conservation" this is what it means, that one cannot apply the conservation sums in the internal lines that hold when balancing input and output lines. They are a mathematical convenience to allow calculation of complicated integrals which the Feynman diagrams describe.
The effects of these particles are measurable, for example, in the effective charge of the electron, different from its "naked" charge.
This is a tautology, because virtual particles in the Feynman diagrams are necessary to make the calculations fit the observations, but one should keep in mind that they are a mathematical convenient representation that cannot be directly measured and thus is not a real situation. One can only measure input and output in an experiment. The calculations fit the experiments well and some people tend to treat virtual particles as something real, which leads to unnecessary confusion on conservation laws.
