Just an update for any curious observers:

I've installed a hotfix because we were experiencing the msiexec issue ( http://support.microsoft.com/kb/2793634 ), so right now (after failover, install, reboot, failback), I have the ERP running on node 2 and training running on node 1 (ie, the reverse of before). I'm going to leave it like this for at least a few days, probably the whole week, and see if I instead see the PLE collapsing on training. If so I'll feel more confident diagnosing it as a potential hardware fault on node 1. If the PLE collapses follow the instance, then that obviously narrows it down to something wrong in the SQL instance itself. If the issue goes away completely... <shrug>... it's amazing what reboots can do...