Thanks for your response.
Unfortunatly this doesn't really narrow it down yet 
The guest OS is custom, with custom LSI drivers. At the time of the crash, there is certainly a spike in disk activity as well as a spike in datastore access latency.
The reason we are investigating this issue, is that the ESX hypervisor isn't virtualizing APIC timer interrupts in a timely enough fashion.
We have a one-shot timer programmed, which is delaying interrupt deliver up to 1 second.
I'm currently correlating this phenomenon with the errors shown in the hypervisor logs.