Have you looked at the logs? It appears that you are using the ASDM for management. Go to Monitoring > Logging > View and see if there is anything useful there?
You mentioned that the topology is three locations all with VPNs between each other. Just brainstorming here but is there a chance that this is a routing issue? For example, you may have a crypto ACL on site A's ASA that sends traffic for site C through site C's tunnel which would be expected. But did you also use the same crypto ACL statements to send traffic destined for site C through site B's tunnel? Perhaps the thought was for redundancy? See the graphic for an example of what I mean:
I feel like this could possibly explain why it works sometimes and other times it wouldn't. For example sometimes the traffic goes the expected route and directly to the intended target site through the most efficient path and other times through the other site (may not even make it at all). I could be completely off base but thought I'd ask anyway.
Another option would be to run a packet capture which can also be easily done through the ASDM. I'd capture on the INSIDE interface and the OUTSIDE interface. On the INSIDE interface capture, filter it down to just remote internal IP's and see if you can spot any high number of transmissions or anything out of the ordinary.
