With this patch parse nodes are allocated sequentially in chunks.  This
reduces fragmentation of the heap and prevents waste at the end of
individually allocated parse nodes.

Saves roughly 20% of RAM during parse stage.