(Though it must be said that a 99.9% uptime service level agreement means they're allowed to be down for 8 hours a year, which some categories of business would consider unacceptable. Still, it's much better than you or I have managed lately.) (And god, how I love Google sometimes. I just typed "0.1 % in hours per year" into the search field and up it comes.)
Well, the last place was down for 8 hours a week so 8 hours a year should be at least a 12-fold improvement. I'm not getting too excited about these people yet - all my previous providers (bar Nik) had certain shortcomings. But I was very impressed with the quick set-up (within an hour of my funds hitting their account) and am comforted by the fact that it's not infeasible to pop round and speak to them in person if things go horribly wrong.
I was disappointed to read that SABRE (the global airline res. system) is to be taken off mainframes running TPF. They used to quote ridiculous donwtime costs (something like USD1m per minute), due to the fact that they handled up to 9000 transactions per second. They're porting the system to Compaq NonStop, of which I (currently) know nothing, but I'll be looking into it. A few thousand TPF apps and ops programmers could be re-training in a few year, methinks.