Tim_one's change to aggressively overallocate nodes when adding child

nodes (in Parser/node.c) resolves the gross memory consumption
exhibited by the EMX runtime on OS/2, so the test should be exercised
on this platform.