And, frankly, 99% is not exactly high-reliability; it represents more than three and a half days downtime per year. If your site is business-critical, then three nines is cutting it close (it's still eight hours and forty-five minutes downtime before the SLA is breached) and would represent a cost compromise. Four nines (just under an hour per year downtime allowed) is expensive for a single server instance; anything more requires a distributed architecture (clustering with failover or multiple app instances with redundant data storage).
It's 99.9, not 99; at 99.9% per month it's 43 minutes a month. That's also assuming it's down at all; the guarantee I think is more for minor outages/updates but I've never had an issue with the site being down more than 20 minutes a month (and that was once for an apache update only). Pretty good for 2 years on paid
Edit: I think my math is off by a pretty hefty amount actually, I think your #'s more accurate, but in any case i've never seen it down more than 20 minutes without the cause being located. The few times it was slow were resolved quickly (usually some moron running RapidLeech and being granted a "goodbye you're suspended").
Edit 2: No, apparently mine is right. 30 days times 24 (hours a day) times 60 (minutes an hour) = 43,200, and 0.1% of that is 43.2 if I'm to trust Google. That'd be 43.2 minutes maximum downtime in 30 days...In any case that is about 8 hours a year, but that's assuming it actually -is- down that much