Experience and engineering nouse. Yes, okay, it is not possible to rule out other causes entirely and I have even seen an un-networked HD-FOX without a drive freeze, but these events are very rare.
We know network events cause crashes, and therefore that the network interface code isn't as well tested or robust as it could be. Why would that be? Because network abilities were an afterthought, and at the time of market not a significant factor. IIRC DNLA was not included in the original firmware roll-out.
We see plenty of reports of disk faults slowing down HDR-FOXes to a crawl, and I have experienced "RC overload" crashing it, but we don't get reports of disk faults resulting in crashes unless there is also RC activity. Conclusion: the disk interface code is well proven and all the execution paths valid (as you would expect from a mature OS kernel).
If the errant execution of code results in the program counter landing on a random (unplanned) memory location, the result will be unexpected actions, lock-up, or system restart. It is not possible to separate lock-ups from restarts, although restarts could also be triggered deliberately if the system is designed to detect faults and reset in response.
So far I have only talked about mis-execution caused by inadequate programming. This is when an external event or exact combination of events has not been anticipated and accounted for in the code or the testing. Programming cannot overcome mis-execution as a result of hardware failure or disturbance - by which I mean the state of a memory cell or register being altered by a radioactive particle (cosmic rays, ambient radiation, radioactive nuclei in the chip packaging), or the functioning being disturbed by signals (voltage, timing) or power rails being out of specification.
The PSU can become marginal (in which case the whole system will be unreliable), there may be mains-born disturbances (failure will be erratic, and likely other equipment will also be affected), the HDD may go out of spec (failures will be persistent).
On balance, the opportunities for network activity to promote crashes overwhelm all other possibilities unless the crashes are frequent and persist with the network disconnected. The very occasional crash could be triggered by a random event, but that's just noise!