As I mentioned before, once the problem is solved that you can load the blockchain while mining, this will prove the client is stable enough to run without crashing.
Okay, that sounds like stability issue 1 . . . it's weird that you'd be seeing problems on all your machines, when many others don't have any stability problem . . . I wonder what might be common to all three machines.
I'm assuming these aren't hard machine crashes, as you know which block you're going down on. It would be helpful to get the tail end of the debug.log file after a crash . . it will probably have crashed somewhere in the middle of the grant allocation code, and then there will be some superfluous network activity . . . just before the network activity is where the interesting stuff is.