we've used 1.41 x 600 time as a predictor of 800 time. so a 600 in 1:18.0 predicts a 1:50.0 while a 1:25.0 would equal 1:59.9. we've never time trialed an all out 600 since i've been here but sometimes people will race it indoors and the formula holds up pretty well. we do have a workout of 600-300-300-150-150 where we never come to a stop for even one instant between reps but have to keep jogging at 45 seconds per 100 for the same distance that we just ran. so the fast 600 rep goes straight into a 600 jog in 4:30, the fast 300s go straight into 300 jogs in 2:15, and so on. all of the reps are supposed to be at 800 pr pace or faster when we're at our peak. we do that workout about twice a year, usually once indoors and once outdoors. when we did it this year 4 guys ran in the 1:21s on the 600 (maybe 1:20-high for the fastest guy), 40-41 on the 300s, and 20-point on the 150s (somebody had too much left and did a low 19 though). all the fast stuff was at 1:47-1:50 800 pace and the same guys all ran 1:50-point or better outdoors, which is just what the workout indicates.
we did a 700 all out time trial once this year and the rule of thumb we're given is to multiply the time by 1.17 to get an 800 time. 1:34.0 predicts a 1:50.0 according to this. i read seb coe did a 700 in 1:26.7, which would predict a 1:41.5 800 and as we know, he ran 1:41.73, so it seems pretty accurate all the way to the world record level. this year on the 700 we had two 1:33-high guys, two 1:35s and two more 1:36s. this predicts two 1:49-highs, two 1:51-lows and two 1:52-mids. The same guys ran seasonal bests of 1:47.9 relay, 1:50.40, 1:50.66, 1:50.80, 1:51.8 relay and 1:52.61. except for that 1:47.9, they were all within half a second of what the 1.17 x 700 time predicted. in 2005 we had two milers run 700 in 1:34.1 and 1:36.0. they never ran the 800 but ran mile times of 4:01.82 and a hand-timed 4:04.5.
so there you have two pretty reliable formulas. multiply the 600 time by 1.41 or the 700 time by 1.17 to predict the 800 time. obviously the 700 is a better predictor since it's almost a full 800.