DV,
Round numbers only: if the normal annual activity rate of a Tornado crew is 200 flying hours per year then the tolerable (according to RA1210) incidence of fatal mid-air collision, based on first-party casualties only, is one fatal collision per 100,000 flying hours (on the assumption that both crew are killed). This is 500 crew-years' worth of activity. If there are 50 Tornado crews at any one time, this is one fatal collision every 10 years.
Even though I've picked deliberately round numbers, you can see that these are much more 'in-the-ballpark' than the millions of hours you quoted. Another point is that Cat A airproxes are not fatal. Not even all mid-air collisions are fatal (I can think of at least 5 off-hand in which all survived, whether through ejection or safely landing the damaged aircraft). So airprox and collision rates don't provide as straightforward an input into the risk calculations as might be expected.