A fever, unless it's incredibly high, should not raise the temperature of the brain. The brain in particular is very sensitive to temperature changes-- all those neurons doing all that continual work creates a lot of heat-- so there's a considerable cooling mechanism in place to prevent that.
The whole "metals melting and creating connections" sounds like hooey. From some of the work I'm starting to do with JNSQ (and a particular computer model I've designed to illustrate autism very simply) it's based on the idea that the underlying neuroanatomy to autism is a greater short-range interconnectivity and a decreased long-range connectivity (both of which have a ratioed relationship to one another, i.e., when there's an increase in short-range connectivity, the long-range fibers decrease and vice versa). If metals created bridges, it wouldn't be selective in just filling in the "gaps" for the long-range white matter tracts-- aside from the issue that it'd be some pretty strange acting metal if it magically knew how to behave like dendritic and axonal arms.
"Oh look kids, the mercury magically knew to take the shape of an axon and to connect this neuron to this other one. That's why little Jimmy's not autistic for the next few days!" Tbh, metal physically connecting two power sources? You'd have seizures and an increase in autism traits, because once a neuron has fired, if it's physically connected to another one (and in normal synaptic relationships, the neurons don't actually touch each other; that only happens in ephaptic connections) there's no inhibition in the world that'll keep that next neuron from firing. The conduction will treat the connected neurons as though they are one, if the conduction is strong enough to travel the entire length of the two neurons that is.
If there is something to the improvement in symptoms during fever (which I'm not inclined to doubt, although this obviously doesn't happen for every autistic person) it's that the immunological reaction somehow alters the usual relationship between excitation (excitatory pyramidal cells) and inhibition (GABA interneurons). My best guess would be that it somehow effects short-range inhibition, increasing it even more. In which case, it doesn't actually reverse autism, but would decrease some of the repetitive firing of neurons, making them less sensitive to stimulation. I.e., it decreases short-range reactivity but doesn't do anything for increasing long-range connectivity.