Thanks again for the ideas Mark.
What you describe is possible. The (small) ISP say that they have primary and backup call providers for their SIP trunks and the issue is that the Cisco UC320 isn't capable of hairpinning the call. Their current temporary fix is to route hairpinned calls in and out via the same provider. When the UC320 was installed, their primary was provider A and failover provider B. For cost reasons they reversed this, and all went to pot. For testing, they reverted back to the original situation (primary A, failover B) and everything worked again. Now we're back to primaryB failoverA so the problem remains.
In terms of the link to the SIP provider (the ISP), there is a single Miktroik router connecting the business to the ISP via a wireless unit. Speeds are generally 50mb down/10-20mb up.
Could be that the packets are going out from one leg and coming back from another and do not know where to go at that point. From what I can imagine, you are having BGP/routing issues here and it's not the SIP server that has the problem. Do you have one or two links to your provider(s)?
Also, is the 'dead air' only one way or both?