Whilst at the Alaska Workshop in Monrovia recently, it was mentioned that if a
BP had an "onEvent" activity, it was not possible to use persistence and
failover for that BP.

I would like to say that the intention of "onEvent" is to accept asynchronous
receives into a BP at any point within the life of the BP, a good example of
this is an asynchronous cancel (in fact this is the example given in section
12.7.1 of the BPEL 2.0 spec). If a BP requires an asynchronous cancel it is
alomost certainly a long-running BP so would almost certainly require
persistence/failover.
The example scenario used at the workshop was based on a real customer scenario
(Ardatis) and they had long-running BPs with asynchronous cancels, just like this.
I think it is imperative we support persistence/failover for BPs with "onEvent".