You guys are nuts - there's no point in sieving at that high level, and you'll break even at best. This is getting extremely close to the point where the average time it takes to find a factor is longer than the time to finish a PRP test.
Unlike GIMPS, we're not trying to find all the primes in the range being sieved. A prime for one of the six k's at 17M, for example, means that all the factors for that k above 17M are now useless, since that k is eliminated. When you take this into account, the optimal sieve depth will be lower: around 50-75P.