Thanks for the replies.
Actually, following some further testing the quality is acceptable apart from if the person on hold is on a mobile phone when it sounds really terrible even with good mobile signal. Maybe it is the mobile network doing something to it?
We find this with several systems that have embedded file based MOH... Avaya IP Office, Mitel 5000, Mitel 3300, ESI Comm Servers, Asterisks/Digium, and others, it is not a "phone system" issue per se, but the cellular network which is not designed to play continuous stream of sound like hold music, it is intended to carry voice which is actually very broken audio when you break it down into tiny bits.