how "off" (as a percentage) would an attempt at geostationary orbit need to be (in either distance or speed) for a satellite to fall to earth in 100, 1000 or 10,000 years? (assumes initial setup only, and no subsequent adjustments).
Being off the nominal geostationary orbital velocity won't cause the satellite to fall to Earth.
For a given altitude, there's one orbital velocity that yields a circular orbit. If your velocity is higher or lower, you instead get an elliptical orbit. As long as that elliptical orbit doesn't intersect the Earth or a substantial amount of the Earth's atmosphere, it will remain stable for quite a while (although, because of disturbances from solar wind, the unevenness of Earth's gravitational field, and the influence of other planets in the solar system, it may not stay up for tens of thousands of years).
Vanguard 1 has remained in orbit for over fifty years, and at the low end of its orbit its altitude is about ~660 km. Even at apogee it's much, much lower than geosynchronous. Vanguard is expected to stay in orbit for 240 years.
Circular orbit velocity for geosynchronous orbit altitude is 3.07 km/s. If you hit that speed, due East, at an altitude of 35,786km above the equator, you're in geostationary orbit. If you're going slower than that, you're in an elliptical orbit, with that point as your apogee.
Let's assume that Klassensat-1 reached geosynchronous altitude by doing a transfer from low Earth orbit but its circularization burn failed, leaving it in a 660km perigee, 35,786 km apogee orbit. This should last at least as long as Vanguard. The speed at apogee for this orbit works out to about 1644 m/s.
So if your orbital velocity is only 54% of geostationary velocity, you'll still stay up for well over 240 years.
If it stays above 900km, your satellite's lifetime will be on the order of 1000 years; this requires only a little more speed at apogee, 1668m/s.