2.0 performance under heavy DGM scenario (load phase)

This bug is cross-related to http://www.couchbase.com/issues/browse/MB-5937

100M items heavy DGM mixed test on vms and physical machines (build 1717 vs 1.8.1-938)

-) 2.0 took more time to load 100M ( 4.07 hr vs 1.34 hr). Disk write queue were constantly higher. (page 4).

+) 2.0 took less time in access phase. (0.8 hr vs 2.8 hr for 1M)

+/-) both 1.8.1 and 2.0 - swap usage didn't go up on the physical machines in access phase, but went up on vms. (100M)

Graphs:
#1 - load phase, physical
#2 - access phase, physical
#3 - load phase, vms
#4 - access phase, vms