Firstly the latest firmware is not 1.9.0, not 1.8.5!
However, I've got 2 WANs load-balanced at firmware 1.9.0 which I set up at 1.8.5 (not via the wizard though) and it always worked perfectly at 1.8.5. It works 'kind-of' perfectly at 1.9.0 but I haven't experienced the problem you have.
The only thing I have seen is that the 'lb-local disable' option has stopped me being able to use 'curl --interface' to go out of the non-default WAN interface, doing a request to icanhazip.com just gives my primary WAN IP address no matter what interface I tell curl to use. Not sure about /bin/ping as it's hard to tell where that is sending its traffic but I guess it's the same thing. Probably all local traffic is now being over-ridden to the primary WAN or something. Not ideal but not a biggie for me.
Otherwise the load-balancing continues to work perfectly from what I can see... traffic is shared and if one interface goes down the other gets all the traffic. When the interface comes back up the load is shared again.
Have you tried creating a 'transition script' to log what is happening? that would let you know if interfaces are going down etc. You could also either set an IP to ping as the 'test' instead of using the standard hostname in order to rule out a problem with DNS lookups causing interfaces to go down (shouldn't happen but you never know, you'll know more if you start logging the transitions).
