Actually, if you look at the number of airline flights that take place in any given year compared to the number of fatal accidents, you'll probably find that the safety rating is somewhere around 99.999%. There are well over a million airline flights in the US alone each year. In a normal year, there might be one or two fatal accidents. Off hand, I remember 5 fatal crashes in the US last year but 4 of them were due to terrorists hijacking the planes and deliberately crashing them. That tends to skew the numbers somewhat.

Rockets are currently nowhere near the reliability of airliners. The Delta II family is one of our most reliable ELVs with two failures in over 20 years (one in 1986 due to an electrical system malfunction and one a couple years ago when a strap-on rocket exploded). It's reliability is somewhere around 98%. The Russian Soyuz booster has a very impressive reliability rating. After a string of failures during development, they actually launched several hundred of them in a row without a single failure. If you look at their launches over the past 20 years, you'll probably find a reliability rating upwards of 99%.

The Shuttle has flown a little over 100 flights with one failure, so the rating is a little over 99% right now. Accepting a 1 in 20 failure rate (95%) like you first suggested for an expensive manned vehicle like the Shuttle is simply unacceptable for anything short of wartime. Even during the height of air operations by the 8th Air Force in WWII, a loss rate over 4% was considered unacceptable and unsustainable.

Your compromise rate of 99.9% allows for one failure per 1,000 flights. That's actually a pretty tough goal given the current state of the technology. One reasons airliners are so reliable is the fact that we aren't dropping pieces off of planes (at least not deliberately) with every flight. Airliners are much less complicated than the Shuttle, too, so there's less stuff to break. Every airliner design has a Minimum Equipment List (MEL) that specifies what must be working and what is allowed to be broken when the flight begins. Right now, we'd never deliberately launch the Shuttle with a broken component.

There are two conflicting approaches to radically lowering launch costs. One approach, as illustrated by the Shuttle and SEI, is to go for the highest level of reusability for spacecraft. The idea is to make spacecraft more like airliners with less equipment thrown away with each mission, less servicing required before flight, and fast turn-around times. Unfortunately, this requires a lot of expensive R&D (X-33, anyone?) which usually means you won't have very many vehicles. Having such a small fleet, you tend to take greater care launching them (kind of like with the B-2 bomber) so the flight rate isn't as high as you'd like and the per-mission costs are pretty high, especially when you amortize the costs of developing and building the spacecraft over their useful lifetimes.

The contrasting approach is more like what the Russians have used successfully for over 40 years - build Big Dumb Rockets in large numbers and launch the hell out of them. At the peak of their space program, the Russians used to launch about 60 Soyuz/Molynia (R-7 family) boosters (almost all carrying military satellite payloads) a year. They literally mass produced the things so there was virtually no R&D costs and everyone knew how to build and launch the things. Big Dumb Boosters aren't as sexy as reusable space planes but it's hard to argue with success.