I've found out what has been causing the server problems all this time! As the failure occured during backup, I monitored the precise time it happens... It is in one part of the DB only.
It appears that the InnoDB tables have become courrpt in one part, which causes MySQL to fail once accessed. However, it isn't easy at all to fix a courrupt InnoDB - so I'm going to process each table individually late at night.
In the mean time, there is an effective fix in place which was tested when the site would have otherwise "crashed" last night. It was down for less than 1 minute
I know, it is strange isn't it! I've read the InnoDB forums and its a known "feature" (not a bug apparently ). If the DB rollback can't fix the row, it just hangs until reset - great I'm going to check for courruption each week and hope that sorts it.
There is now a connection fix and DB courruption reset proceedure in place should either of the 2 problems that sometimes occur happen. Lets hope thats all it needs